{"_id":"@alainabbasi/atlas","_rev":"18-171d57337157db32fbecb2806947f85b","name":"@alainabbasi/atlas","dist-tags":{"latest":"0.7.0"},"versions":{"0.1.0":{"name":"@alainabbasi/atlas","version":"0.1.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"64f43b9de4c5aaa95a2ed31fddc3475ab8454272","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.0.tgz","fileCount":26,"integrity":"sha512-vd+cevG/h3PlqEo8GZ2NWtT0rLzSoeDhaSfX0MHceSHTZIIMGkkaOHaUMVNdz25O5UzorlrCFTAlj5V5YeT7lw==","signatures":[{"sig":"MEYCIQCYwUlpaouAoIDdaInr+Qy0gthD05Inv8nxj/mtSMnbIgIhAK6Az4+YCuau7262yuyXYJtD/fPdeMNugg6DYXvA1lnl","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"unpackedSize":165877},"engines":{"node":">=14"},"gitHead":"215acb35d55c0d462bcdbc85b887a9985379b830","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.2","description":"ATLAS — Agentic Harness Standard. 10–30× fewer orientation tokens. Zero infrastructure. Three Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"23.11.1","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.0_1780462375759_0.38090453539027913","host":"s3://npm-registry-packages-npm-production"}},"0.1.2":{"name":"@alainabbasi/atlas","version":"0.1.2","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.2","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"0914dad270c34aaf1ca23ea539275e7a3a58a8e5","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.2.tgz","fileCount":26,"integrity":"sha512-efE/aSUCSbyAae/cMOAYZKLuY/egMx3YkVeKG7nJhqE0124HVCSMtCUAyXQy66AFuKzo9fTX8vct9VPyvlJiPQ==","signatures":[{"sig":"MEQCIH2BG3ST3VozAM1rUJq41xAaLBDzTWcYWFQ+veikD0U5AiBb35I+cAOYnrRqUKCo2MfIo7gKXdsepwBsiM6nCc3qSQ==","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.2","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":186874},"engines":{"node":">=14"},"gitHead":"93b427ace10e4c966dde7833de145830b7ab8077","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. 10–30× fewer orientation tokens. Zero infrastructure. Three Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.2_1780876624586_0.41473133829078","host":"s3://npm-registry-packages-npm-production"}},"0.1.3":{"name":"@alainabbasi/atlas","version":"0.1.3","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.3","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"34173eff4fda7557ccb0c31968e37695687c5c40","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.3.tgz","fileCount":26,"integrity":"sha512-b4dTu1ICeNp9Dg8g/3Zz9cNxwQPYX3HNsrIBSKAzZv0t+w5SYpzMJRJykdWRdqGmj6dLzL3nz2NtVWvqACdxcg==","signatures":[{"sig":"MEUCIQC0DnkiCodBYLBWW0XUuzETcGeXRnVdRbwldtTb+D2SAQIgTuOXv1NrY3MlAgs/Q4Zxw/QEmXmnh+Kvj3D7gOC6g7Y=","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.3","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":186874},"engines":{"node":">=14"},"gitHead":"87b65c2b2d50193469b771b8c889e8a321383c9c","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. 10–30× fewer orientation tokens. Zero infrastructure. Three Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.3_1780877757612_0.02507560306112766","host":"s3://npm-registry-packages-npm-production"}},"0.1.4":{"name":"@alainabbasi/atlas","version":"0.1.4","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.4","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"f3feba19441123aa91f775e45a6618a3235d614c","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.4.tgz","fileCount":27,"integrity":"sha512-TosFbAZGwhCOCsWHv3Ky2ameh+uhOLC4yzeey4vdp7ZAbBeOAtsAgRDBkyvWIm5hmG2JGhXvsrUo+3oeRt3gYA==","signatures":[{"sig":"MEQCIGRCUBOV4pcD8lrLEGqy4x7FVLLFleIJtV7FxMhEVhFsAiAnt8dBCPH3n6GFBSuIqs/qfUaDaUAPvOhLU00bCqWh2Q==","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.4","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":196965},"engines":{"node":">=14"},"gitHead":"3109067069f8d16801140f85894d4d671df1cb6e","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. 10–30× fewer orientation tokens. Zero infrastructure. Three Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.4_1780881322013_0.17853787472972105","host":"s3://npm-registry-packages-npm-production"}},"0.1.5":{"name":"@alainabbasi/atlas","version":"0.1.5","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.5","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"2302fb779515f900ea93bb5bc09e857e2301ea9b","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.5.tgz","fileCount":27,"integrity":"sha512-NqSk+2OJ54D9nbTwiv5v0HFgunfbrQTuA9/OHpyN5l2RyCZuaEMCEJ3b3smfcXYH/98ugbm6zQ+qCRDiuEUBEw==","signatures":[{"sig":"MEUCIQCZoi2wd9g+TX2kHd4nc9uLHgSfjjAIsG9wd5ek436ZkQIgDrGP/1POYpTFCpukVIfkuHfr4jMGgK13x/qtDfiPHg8=","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.5","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":210312},"engines":{"node":">=14"},"gitHead":"75c79d5e3796c1e7f84c1e325603939cf016fbc5","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.5_1780884989085_0.023596995806444143","host":"s3://npm-registry-packages-npm-production"}},"0.1.6":{"name":"@alainabbasi/atlas","version":"0.1.6","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.6","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"ab05aae3adec3e8b3164b4bf80d8842b3687a1e4","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.6.tgz","fileCount":27,"integrity":"sha512-IY+utPn6poOcyxR2Ee2MZVVX5dp2iJXjBogwQj+ALgiIuYdL1JZHhw8MZznOWDhjpaFt6nfWimAE2kc4ZjnejQ==","signatures":[{"sig":"MEQCICViSuqLE8rX4sP+M3NpWfE08lXHdgrj80/0bGRtleIPAiA8jQavwddHWqWAu8KU4GTDdU3Tr2/ge//tNlC+9Zy5/Q==","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.6","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":228252},"engines":{"node":">=14"},"gitHead":"fffc4ecb0db9d7f1fd3abd00579a895b38981c79","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.6_1780936208688_0.07875152565948973","host":"s3://npm-registry-packages-npm-production"}},"0.1.7":{"name":"@alainabbasi/atlas","version":"0.1.7","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.7","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"b1b75c9f1f835e2ef3a3167a8a2a03e8c8da1912","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.7.tgz","fileCount":28,"integrity":"sha512-JjjRtIw+f/1RXF22FECRCGvBcWAZ4xZYgGiyBulObgK41QOdi5nFnzs/wZVW7s2ZUB9j51LQi78/juwZJtW9gQ==","signatures":[{"sig":"MEYCIQD10WZkX1C8i0ARwnPKycoc6zp271HnsUjMu52V8o3g8wIhANGpjjeuqgmyY8TutHb/Tk3LfjdfXTkfkVO8TRpRp5pV","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.7","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":244101},"engines":{"node":">=14"},"gitHead":"1b35171b105c5605bcbe9f410748c5ff37b64fb2","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.7_1780960811926_0.05849364226977638","host":"s3://npm-registry-packages-npm-production"}},"0.1.8":{"name":"@alainabbasi/atlas","version":"0.1.8","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.8","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"fa046d35c2391e5e2eddf40e5f6701f17d5380f0","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.8.tgz","fileCount":28,"integrity":"sha512-GxcKLzh2sNfDovyaHui3VwGHiaHbBmJtHM7OhgzX4azEFLIThujnF2lYWfPywd2sF+nCpaJsQCmcLFrizoJHuA==","signatures":[{"sig":"MEYCIQDxNLe25UUt0k/Gz3ZhDrbCtP88PlSG+V7AnOd+wFAK8QIhAIYMHEUMdJf648EvZHXMY5iIaCtbQCkvuMk45VSshqym","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.8","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":252353},"engines":{"node":">=14"},"gitHead":"a557e1806ecd400030b44e8b1487b0cbcc6fa1ef","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.8_1780969252701_0.4058893906332961","host":"s3://npm-registry-packages-npm-production"}},"0.1.9":{"name":"@alainabbasi/atlas","version":"0.1.9","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.9","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"406900ccc5127632808980b9cf116b976b237cd7","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.9.tgz","fileCount":28,"integrity":"sha512-YHOEZU6Xnuw0aZsVSA1QJWKXpuj0kv6sKXxMPaW93uRXoOtsEDAuoTDU6NGVgHyk3AAix8AC8LvaVeJYqTIJ2A==","signatures":[{"sig":"MEUCIQCSapAWHyWehozy22W8rNECNUysixpFDeyXOEqJG/jsbQIgSbFRklaRxEeJXXi8GRc8imXZm6aEi8zKCR+ElYuskyM=","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.9","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":258249},"engines":{"node":">=14"},"gitHead":"656d5cb893671da2019d0fe6ca2b34b9ab03f758","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.9_1780970960224_0.9426404093500971","host":"s3://npm-registry-packages-npm-production"}},"0.1.10":{"name":"@alainabbasi/atlas","version":"0.1.10","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.10","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"79e74e3158473cd508ca7f3f7579d91bb78e506b","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.10.tgz","fileCount":28,"integrity":"sha512-gSqhNKMkWZjFGk2PQ2/yc7T0KOkJgJAVa0amGKIT0QToRlJcQ85EbEMtViv6bgCyKWZvoJdpIb4wni9gfUIucQ==","signatures":[{"sig":"MEUCIQDr08CrmRLh5O1wnEgmK3fu8luZjQk2rSZQYemFlg9caAIgeRZfVKgXoWzMNC2+2MWuyz9OKq1UQMDxK0gIkeoSbyM=","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.10","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":263196},"engines":{"node":">=14"},"gitHead":"fb1f570c11028ee442533b14c66d7c8a9b7c896f","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.10_1780974530702_0.3374605293669759","host":"s3://npm-registry-packages-npm-production"}},"0.1.11":{"name":"@alainabbasi/atlas","version":"0.1.11","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.1.11","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"57a82262b4f43e9eab1599f1b17b88b4561e467f","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.1.11.tgz","fileCount":28,"integrity":"sha512-ZAMb4CjSCtAMg+l3LYC/6owNLtZ3EMGAgEa65KRbct0FYO/UlHutvGBYP1Ez9UX0f3FhC1/T5DDeFZDcig+Kqg==","signatures":[{"sig":"MEYCIQDQZ54w/6dx9iwWsiRk9LHlgMxNFkWle4Yw8nkG9vk8FgIhAJ6DqPPCEXRMzPTRHDZe3oLkQ8Qram/EqkxCdfNLnM21","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.1.11","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":265142},"engines":{"node":">=14"},"gitHead":"034dc9e058108a5e0681c336671a061aef3930e5","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.1.11_1780976420695_0.9757628290912115","host":"s3://npm-registry-packages-npm-production"}},"0.2.0":{"name":"@alainabbasi/atlas","version":"0.2.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.2.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"bb09b53769453f4e98bc5329fb968eb296ad5522","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.2.0.tgz","fileCount":28,"integrity":"sha512-gP0aAIJl15NZKfXCZina2azxm19X1bnQ49IBzQKW7ycNIKrezlUyhWyRQunxvcDTcen1udsLuaU/469kjqD+Dg==","signatures":[{"sig":"MEYCIQCuLRlfbL+FzP7QlTKcdE5jtsNSwdUcdf8gJHvlOKm7fAIhANu86ozcQKD5fvH9kDRYJ0IE/8/cLLcGsRvE+76AJH1D","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.2.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":276989},"engines":{"node":">=14"},"gitHead":"c134d0c75c2db7a69b48920702acdeb317263543","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.2.0_1781896242838_0.07698107266047671","host":"s3://npm-registry-packages-npm-production"}},"0.3.0":{"name":"@alainabbasi/atlas","version":"0.3.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.3.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"0863f6b6e7785ea9d225469d4d8c8cf4086989eb","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.3.0.tgz","fileCount":28,"integrity":"sha512-5nLl6w8N9j74tHjmBCtfUzavhVnzw+o2V2b2BXBsVLxc19mxmK4lsyhhxiOs0bM97jP/qhn8BuHshXZsbxAo4A==","signatures":[{"sig":"MEYCIQCXnwhX/zbJqFaqW0kOuPkbcm1ahVSkbReZpivCMJMwjgIhAIJ8uKOuju9EjdoQeT4XYbkJmOpxcFMSGa4TMbMDKDUK","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.3.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":284034},"engines":{"node":">=14"},"gitHead":"ae92238102dee19e79b89ee5d9624c34ecd484df","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.3.0_1781898998835_0.09039180159261884","host":"s3://npm-registry-packages-npm-production"}},"0.4.0":{"name":"@alainabbasi/atlas","version":"0.4.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.4.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"f9876c611eaf493786c69fb2441a84b26d63b5c6","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.4.0.tgz","fileCount":30,"integrity":"sha512-JET+JAy3YYjWvq6QyN+uG92mRRUeyDB2p85jkxD8g6Q8Gw4ZXgWhXS6YHJMpPKbuMktLcDvXkDGVxumMU6MnYA==","signatures":[{"sig":"MEQCICgoVgC0R7iwagnNpl5NXDvRffPJODN/HgzVy8u3LO47AiBKiVplfuJo94tmw6TCPgKo2XkOpelwkIXkyDMRWw8AHA==","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.4.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":290660},"engines":{"node":">=14"},"gitHead":"e859de80b70b77e9d7c699ed96da9bc76d5f712b","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.4.0_1781915362675_0.7799785369722829","host":"s3://npm-registry-packages-npm-production"}},"0.4.1":{"name":"@alainabbasi/atlas","version":"0.4.1","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.4.1","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"5235aa44e334a6c3ef02dd339ad33b52d06c267b","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.4.1.tgz","fileCount":30,"integrity":"sha512-MslOVUDzpdHyrp/p3L3HzbruEfa2ZFHuk+0D/0a3iDgEAN8RMx+LbDdwOymnCHIYQcTTXGZ7bI63Q0eKn7dU+A==","signatures":[{"sig":"MEQCIA5SstfzXcOzcCs76dGrmpx2mOjEBukHn9Qe0K9vbwNWAiANMSVpue/O3cz446ZYmolW46xeTVBSvSH8ENls5TDGbg==","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.4.1","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":291986},"engines":{"node":">=14"},"gitHead":"d91f45da0cd603490f96a679aa5742549fa30bef","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.4.1_1781916749736_0.37008611101124633","host":"s3://npm-registry-packages-npm-production"}},"0.5.0":{"name":"@alainabbasi/atlas","version":"0.5.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.5.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"2a099143c4148f922e4711e33932fe431968c69a","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.5.0.tgz","fileCount":30,"integrity":"sha512-pcpj4dPw0h99lUh7xF/jn+LSVHAo1GZxgAH9NP1uqMsbNohnVSq4mHre3sBdOOmAHWy2+Jk1Hb9QWdh9tiI2WQ==","signatures":[{"sig":"MEUCIQCYoJR2uix4bMkztbgN3FB46NX6WjC9CsYX97iipsDBGwIgRp7W0QB5D+M3YYxkRDvQD3nLHMEnO3lUS53bY+xK/d4=","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.5.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":295430},"engines":{"node":">=14"},"gitHead":"390f635190598dafc7aecc831476e9dde0ec47c7","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.22.3","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.5.0_1781918514982_0.1984959929430885","host":"s3://npm-registry-packages-npm-production"}},"0.6.0":{"name":"@alainabbasi/atlas","version":"0.6.0","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"author":{"name":"Alain Abbasi"},"license":"MIT","_id":"@alainabbasi/atlas@0.6.0","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"homepage":"https://github.com/Abbasi-Alain/atlas#readme","bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"bin":{"atlas":"bin/atlas-node"},"dist":{"shasum":"dfe01bb966a81cde9b2ee35d3d0a0468074a258b","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.6.0.tgz","fileCount":36,"integrity":"sha512-sUKHl/T8AmCAyxyUTgx0TtWGkI0H0ueVknpysDK4PB/uqrDK2p3b/5EUsqTFSJ2upDMj8HpHp+I3GKfKJhajnQ==","signatures":[{"sig":"MEYCIQDjmEd4bIooYm1X++Cnf016DYF8AhGNoWQ1uvPvya6znwIhAIz7PXwP+w09tpC0ZHKTL4RkAzo9phHUDa2gwBS2Mo7p","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.6.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"unpackedSize":361918},"engines":{"node":">=14"},"gitHead":"314a30f752098f3d58457ae27cf3da0865d89308","scripts":{"test":"bash tests/bootstrap.test.sh"},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"repository":{"url":"git+https://github.com/Abbasi-Alain/atlas.git","type":"git"},"_npmVersion":"10.9.8","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","directories":{},"_nodeVersion":"22.23.1","publishConfig":{"access":"public"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/atlas_0.6.0_1783736683467_0.937867107978027","host":"s3://npm-registry-packages-npm-production"}},"0.7.0":{"name":"@alainabbasi/atlas","version":"0.7.0","description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","license":"MIT","homepage":"https://github.com/Abbasi-Alain/atlas#readme","repository":{"type":"git","url":"git+https://github.com/Abbasi-Alain/atlas.git"},"bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"author":{"name":"Alain Abbasi"},"bin":{"atlas":"bin/atlas-node"},"keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"engines":{"node":">=14"},"publishConfig":{"access":"public"},"scripts":{"test":"bash tests/bootstrap.test.sh"},"_id":"@alainabbasi/atlas@0.7.0","gitHead":"bff592eb480935b02643688898708957f59516c5","_nodeVersion":"22.23.1","_npmVersion":"10.9.8","dist":{"integrity":"sha512-1Unu75l2hj00kMpFlx01PhHqYne8K0rzhkOfgd0ofKYjokfGCPDXZyhhlOB8eOh5ylBtVlE84YcgmBkyskAmsA==","shasum":"7cc8a91fd9dff4964dafd38027cfe2ccb9c15778","tarball":"https://registry.npmjs.org/@alainabbasi/atlas/-/atlas-0.7.0.tgz","fileCount":39,"unpackedSize":463209,"attestations":{"url":"https://registry.npmjs.org/-/npm/v1/attestations/@alainabbasi%2fatlas@0.7.0","provenance":{"predicateType":"https://slsa.dev/provenance/v1"}},"signatures":[{"keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U","sig":"MEUCIEtSNiRKm0cMJuc7FtNg8nHPv8QWzSh3WPvJ1RlgYq5wAiEAq/YT3YdIqDjp38tniQdZ/hOWXXT9UQ73R7+3L6WHklo="}]},"_npmUser":{"name":"alainabbasi","email":"abbasi.alain@gmail.com"},"directories":{},"maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"_npmOperationalInternal":{"host":"s3://npm-registry-packages-npm-production","tmp":"tmp/atlas_0.7.0_1785021138329_0.8162769260583858"},"_hasShrinkwrap":false}},"time":{"created":"2026-06-03T04:52:55.651Z","modified":"2026-07-25T23:12:18.845Z","0.1.0":"2026-06-03T04:52:55.901Z","0.1.2":"2026-06-07T23:57:04.729Z","0.1.3":"2026-06-08T00:15:57.798Z","0.1.4":"2026-06-08T01:15:22.149Z","0.1.5":"2026-06-08T02:16:29.217Z","0.1.6":"2026-06-08T16:30:08.868Z","0.1.7":"2026-06-08T23:20:12.082Z","0.1.8":"2026-06-09T01:40:52.874Z","0.1.9":"2026-06-09T02:09:20.360Z","0.1.10":"2026-06-09T03:08:50.886Z","0.1.11":"2026-06-09T03:40:20.834Z","0.2.0":"2026-06-19T19:10:42.993Z","0.3.0":"2026-06-19T19:56:39.016Z","0.4.0":"2026-06-20T00:29:22.831Z","0.4.1":"2026-06-20T00:52:29.864Z","0.5.0":"2026-06-20T01:21:55.128Z","0.6.0":"2026-07-11T02:24:43.597Z","0.7.0":"2026-07-25T23:12:18.504Z"},"bugs":{"url":"https://github.com/Abbasi-Alain/atlas/issues"},"author":{"name":"Alain Abbasi"},"license":"MIT","homepage":"https://github.com/Abbasi-Alain/atlas#readme","keywords":["agentic-harness","ai-agents","claude-code","codex","cursor","gemini","copilot","context-reduction","token-optimization","multi-agent","llm","documentation-standard","project-map","skill-anchor","architecture-decision-records","atlas","developer-tools"],"repository":{"type":"git","url":"git+https://github.com/Abbasi-Alain/atlas.git"},"description":"ATLAS — Agentic Harness Standard. ~92% fewer agent orientation tokens (measured, reproducible). Zero infrastructure. Four Markdown files every Claude/Codex/Cursor/Gemini/Zed/OpenCode/Copilot/Hermes session reads first.","maintainers":[{"name":"alainabbasi","email":"abbasi.alain@gmail.com"}],"readme":"# ATLAS — Agentic Harness Standard — for better multi agents performance, agents token and context reduction by design, graph based. Zero infrastructure. \n\n> ## −92% agent orientation tokens (12.8×) — *measured, not claimed*. Up to −99% on whole-repo loads; bigger on larger repos.\n>\n> Stop your agents from burning **100k+ tokens per session** on `ls`, `find`, `grep`, and `git ls-files` just to figure out where things live. **Four Markdown files** ship the project map, the task playbook, the scar memory, and the behavior contract into context **before the first tool call** — every Claude / Codex / Cursor / Gemini / Zed / OpenCode / Copilot / Hermes session.\n>\n> Not a claim — **[benchmarked + logged](docs/benchmarks/RESULTS.md)**: 12.8× (92%) on this very repo, reproducible with `atlas bench`.\n>\n> **Don't grep. Don't guess. Don't repeat.**\n\n

\n \"ATLAS\"\n

\n\n

\n \"92%\n \"reproducible\n \"install\n

\n\n

\n \"atlas\n

\n\n

\n \"Stars\"/\n \"Forks\"/\n \"Contributors\"/\n \"License:\n \"CI\"/\n \"DOI\"/\n

\n\n

\n \"npm\"/\n \"npm\n \"Spec\n \"Zero\n \"9\n \"Harness\n \"Bash\"/\n \"PRs\n

\n\n---\n\n## The 30-second pitch\n\n\n| Metric | Without ATLAS | With ATLAS | Reduction |\n| ----------------------------------------- | ----------------- | --------------------------- | ---------- |\n| **Orientation tokens per session** | 55–180k | ~1.8k auto + 1–5k on demand | **12.8× [measured](docs/benchmarks/RESULTS.md)** |\n| **Orientation tool calls** | 10–30 | **0** | ∞× |\n| **Same bug re-introduced** | every fresh agent | never (cite `§ANCHOR`) | — |\n| **\"While I'm here\" cleanup churn in PRs** | high | 30–60% lower *(est.)* | — |\n| **Infrastructure required** | — | **none** (4 Markdown files) | — |\n| **Runtimes supported** | varies | **9** out of the box | — |\n\n\nATLAS is **a harness performance tool disguised as documentation**. Three files. No daemon, no DB, no service to run, no SDK to wire. Works with whatever agent runtime your team picked, and the next one.\n\n---\n\n## Why now\n\nThree things are exploding at once:\n\n1. **Context windows are growing, but tokens-per-task keep rising faster.** A 1M-token context isn't the answer if every fresh agent spends 100k of it on `ls` / `find` / `grep` before doing real work.\n2. **Multi-agent dispatch is becoming default.** Lead agents spawn sub-agents (Claude Code's Task tool, Cursor's Background Agents, Codex worker mode, ECC squads, Manus, Replit Agent). Each sub-agent re-pays the orientation tax. ATLAS amortises it.\n3. **AGENTS.md is becoming a real standard.** Codex / OpenCode / Cursor / Gemini / Zed all read AGENTS.md-style files at session start. ATLAS gives that file the shape that maximizes its value — not just behavior rules, but a real map.\n\n## How it works *(in 30 seconds)*\n\n```mermaid\nflowchart LR\n A[Fresh agent session] --> B{ATLAS.md
exists?}\n B -- no --> C[ls · find · grep · read x5
~55-180k tokens
10-30 tool calls]\n C --> Z[Start work
tired + confused]\n B -- yes --> D[SessionStart hook
auto-prints §0 + §1 + SKILL ToC
~1.8k tokens
0 tool calls]\n D --> E[Read one specific module
1-5k tokens
1-2 tool calls]\n E --> F[Start work
oriented + with the playbook]\n```\n\nThe hook fires once per session, before the agent's first action.\n\n---\n\n## The quartet *(four files. that's it.)*\n\n\n| File | Axis | Question it answers | Token cost |\n| ----------------------------------------------------------------- | -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------- |\n| `[ATLAS.md](templates/ATLAS.md.tmpl)` | **Structural** | *Where is X?* Graph index — every important module, its role, what it talks to. Plus glossary, data model, external deps, runtime topology, observability, security boundaries, build/deploy. | ~1.4k auto-loaded §0+§1 |\n| `[SCARS.md](templates/SCARS.md.tmpl)` | **Failure memory** | *What did we learn the hard way?* Stable anchors (`§NAME-LIKE-THIS`) — symptom → root cause → do NOT → do → file pointer. Cite from PRs and commits. | ~0.4k ToC auto-loaded |\n| `[.agents/skill//SKILL.md](templates/SKILL.md.tmpl)` | **Procedural** | *How do I do X here?* Task recipes — goal → steps → how to verify. | on-demand |\n| `[CLAUDE.md](templates/CLAUDE.md.tmpl)` + `[AGENTS.md](#)` mirror | **Behavioral** | *How should the agent act?* Don't assume. Don't refactor adjacent code. Match existing style. Define success criteria. Loop until verified. | on-demand |\n\n\nPlus `[EXAMPLES.md](templates/EXAMPLES.md.tmpl)` — vague→concrete transformations that teach the patterns by contrast.\n\n**Net:** ~1.8k tokens of project orientation lands in the agent's context **before the first tool call**, eliminating 10–30 redundant `ls`/`find`/`grep`/`read` round trips per session.\n\n---\n\n## Benchmarks — measured, not just claimed\n\nATLAS ships an A/B benchmark (`atlas bench`) and appends every run to a\n[ledger](docs/benchmarks/RESULTS.md) so the reduction is trackable across\nversions, models, and dates. First result on **this repo** (deterministic, reproducible) —\nthe reduction is a *range*, set by how the agent would otherwise orient:\n\n| \"Without ATLAS\" orientation | tokens | vs the spine |\n|---|---|---|\n| Whole-repo dump (RAG / \"load everything\") | 104,390 | **−99% · 80×** |\n| Smart skim (README + file tree + source heads) | 16,571 | **−92% · 12.8×** |\n| ATLAS §0–1 spine (auto-injected before the first tool call) | 1,297 | — |\n\n`atlas measure` prints this range for any repo; the `atlas bench` ledger logs the\n**−92%** (the conservative, fair-fight baseline — not the whole-repo strawman).\n\n```bash\n# any OpenAI-compatible endpoint (incl. a local vLLM), free + deterministic:\natlas bench --runtime openai --api-base http://localhost:8000/v1 --model \n# or a live agentic A/B with a real coding agent:\natlas bench --runtime claude # or codex | opencode\n```\n\nThe skim grows with codebase size while the spine stays ~constant, so on large\nrepos even the conservative (vs-skim) number climbs toward that upper bound. In\n**multi-agent** setups it compounds — every sub-agent pays the orientation tax\nindependently, and ATLAS pays it once into the shared quartet. Methodology +\nhonesty rules:\n[docs/benchmarks/methodology.md](docs/benchmarks/methodology.md).\n\n### 🚩 Not just on this repo — measured on the ones you know\n\nThe same deterministic `atlas measure`, after a one-command `atlas onboard`, on famous repos:\n\n| | openclaw | graphify | ECC | hermes-agent | fastapi | django | curl |\n|---|---|---|---|---|---|---|---|\n| **orientation cut** | **−94%** | −93% | −92% | −82% | −89% | −78% | −78% |\n\n**−75% to −94% vs a smart skim**, across the AI-agent ecosystem *and* general OSS — free + reproducible. Full table + how to reproduce: **[docs/FLAGSHIP.md](docs/FLAGSHIP.md)**.\n\n---\n\n## Install — pick your flavor\n\n### One-liner (bash, any platform)\n\n```bash\ncurl -fsSL https://raw.githubusercontent.com/Abbasi-Alain/atlas/main/install.sh | bash\n```\n\n### Homebrew *(macOS / Linuxbrew)*\n\n```bash\nbrew tap Abbasi-Alain/atlas\nbrew install atlas\n```\n\n### npx *(zero-install, Node already on PATH)*\n\n```bash\nnpx @alainabbasi/atlas init\n```\n\n### Debian / Ubuntu / Mint / Pop!_OS\n\n**`apt install` via the PPA** *(recommended — tracks releases, auto-updates):*\n\n```bash\nsudo add-apt-repository ppa:alainabbasi/atlas\nsudo apt update && sudo apt install atlas\n```\n\n**Or a one-off `.deb`** from the [latest release](https://github.com/Abbasi-Alain/atlas/releases/latest):\n\n```bash\ncurl -L https://github.com/Abbasi-Alain/atlas/releases/latest/download/atlas_0.1.5_all.deb \\\n -o /tmp/atlas.deb && sudo apt install /tmp/atlas.deb\n```\n\n### Arch / Manjaro *(AUR)*\n\n```bash\nyay -S atlas # or: paru -S atlas\n```\n\n### Manual clone\n\n```bash\ngit clone https://github.com/Abbasi-Alain/atlas.git ~/.atlas\n~/.atlas/install.sh\n```\n\nAfter install you have the `atlas` CLI on PATH and adapters ready to wire into your agent runtime.\n\n### Update\n\n```bash\nnpm update -g @alainabbasi/atlas # npm\nbrew upgrade atlas # Homebrew\nsudo apt update && sudo apt upgrade atlas # apt (PPA)\n```\n\n> **Cut your agent's orientation cost too?** ⭐ [Star the repo](https://github.com/Abbasi-Alain/atlas), run `atlas measure --share` to add yours to the [leaderboard](docs/LEADERBOARD.md), and see the [measured proof on famous repos](docs/FLAGSHIP.md) (−75% to −94%).\n\n### Verify a release *(supply-chain)*\n\nEvery release ships **signed [SLSA build provenance](https://slsa.dev)** (keyless, via Sigstore) plus a `checksums.txt`, and the npm package is published with **npm provenance**. Verify before you trust:\n\n```bash\n# GitHub release tarball — provenance + signature (needs gh ≥ 2.49):\ngh attestation verify atlas-X.Y.Z.tar.gz --repo Abbasi-Alain/atlas\nsha256sum -c checksums.txt # integrity\n\n# npm — published with provenance:\nnpm audit signatures # after install, or:\nnpm view @alainabbasi/atlas dist.attestations\n```\n\nA failed verification means the artifact was **not** built by this repo's release workflow — don't run it.\n\n---\n\n## 60-second start\n\n```bash\ncd your-project/\n\natlas init # write ATLAS.md + SKILL.md + SCARS.md\n # + CLAUDE.md + AGENTS.md mirror + EXAMPLES.md\n\natlas check # verify anchors unique, structure valid\natlas measure # estimate the orientation-token savings\n\natlas install --runtime claude-code # SessionStart hook + /init-atlas command\n # (or: codex / opencode / hermes / generic)\n```\n\nOpen a new agent session in any project with these files. The agent's first action will be reading `ATLAS.md §0` and the `SKILL.md` table of contents — automatically, no prompt engineering required.\n\n---\n\n## Plug into any agent — the MCP server\n\nATLAS ships a **zero-dependency Model Context Protocol server**, so *any* MCP client — Claude Code, Cursor, OpenClaw, NemoClaw, Codex, Gemini — reads the map directly instead of grepping:\n\n```bash\natlas mcp --config # prints the registration snippet\nclaude mcp add atlas -- atlas mcp # or paste the snippet into .mcp.json / Cursor\n```\n\n**One snippet, every platform — and there's no per-platform code to write.** MCP is a\n*standard*: every MCP-native runtime loads the **same** `mcpServers` block. The server is\nconformance-tested (protocol `2024-11-05`; tools `atlas_orient · atlas_find · atlas_scars ·\natlas_measure`), so if a platform speaks MCP, ATLAS already works in it:\n\n| Platform | Drop the `mcpServers` block into… |\n|---|---|\n| **Claude Code** | `claude mcp add atlas -- atlas mcp` · or project `.mcp.json` |\n| **Cursor · Windsurf · Zed** | `~/.cursor/mcp.json` · `windsurf/mcp_config.json` · `settings.json → context_servers` |\n| **OpenClaw · opencrust** | the platform's MCP registry / config file |\n| **NVIDIA NemoClaw** | the OpenShell MCP config |\n| **zeroclaw · Qwen** (qwen-code) | `mcpServers` + the auto-generated `AGENTS.md` — a fully local, free stack |\n| **Codex · Gemini · any MCP client** | paste the `atlas mcp --config` block verbatim |\n\n→ [full matrix + exact paths](docs/INTEGRATIONS.md#-the-30-second-way--mcp-works-in-every-modern-agent).\n\nFour local tools — no infra, no embeddings, no API key:\n\n| tool | what the agent gets |\n|---|---|\n| `atlas_orient` | the §0 map + SKILL/SCARS tables of contents — *\"where things live\"* |\n| `atlas_find` | where a thing lives, by keyword |\n| `atlas_scars` | the failure lessons for a topic |\n| `atlas_measure` | the orientation-token reduction for this repo |\n\nNeed deep queries? Set `ATLAS_MCP_BACKEND_URL` and three more tools (`atlas_graph`, `atlas_deepsearch`, `atlas_recall`) light up, proxying to a bring-your-own graph+vector backend. With nothing set, ATLAS stays **100% local and free**.\n\nSharing with a team? `atlas mcp --http --token \"$SECRET\"` serves the same tools over HTTP — token auth is opt-in (no token locally, required off-localhost), with `GET /health` for liveness.\n\n---\n\n## Superpowers\n\n

\n \"atlas\n
atlas orient \"<task>\" — the relevant slice of the map + the SCARS that bite this task.\n

\n\n

\n \"atlas\n
atlas map — a Unicode graph in the terminal; pipe it to a .md for GitHub-rendered Mermaid.\n

\n\n- **Task-aware orientation** — `atlas orient \"fix the failing auth test\"` returns *only* the relevant slice of the map **plus the SCARS that bite that task** — not the whole file. Over MCP too (`atlas_orient(task)`). The agent's first move, conditioned on the job. Nobody else does this.\n- **A picture of your repo** — `atlas map` draws a **Unicode graph right in your terminal**; pipe it to a file (`atlas map > map.md`) and it's a **Mermaid** diagram that renders on GitHub, or `atlas map --html` for a standalone page to screenshot.\n- **The ATLAS PR bot** — drop `uses: Abbasi-Alain/atlas@v1` into a workflow and every PR gets a sticky comment with the measured **−92→99% orientation savings** + map-drift status. Zero hosting — runs in your CI. *(This repo runs it on its own PRs.)*\n- **A map that never goes stale** — `atlas hooks install --auto` adds a git pre-commit hook that auto-refreshes `ATLAS.md §0.5` on structural change and stages it *into the commit*. Docs-as-code that can't rot.\n- **ATLAS conducts the ecosystem** — the MCP deep tools (`atlas_graph`, `atlas_deepsearch`) light up when **graphify** or **CodeGraphContext** is installed and route the query to it. Orient free via ATLAS, drill down via whatever you've got.\n- **Leaderboard** — `atlas measure --share` prints your savings + a one-click submit link. See the [board](docs/LEADERBOARD.md).\n- **Onboard any repo in one command** — `atlas onboard` scaffolds the quartet, **auto-drafts a real map** (a §0 \"where to look\" table + a §1 dependency graph, not a blank page), and measures the savings; `--pr` opens the pull request for you.\n\n---\n\n## Style presets — pick a vibe\n\nNot every project wants the same temperament. `atlas init --style ` picks the templates:\n\n\n| Style | When to use |\n| ---------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| `default` | The complete universal scaffolding — §0-§9 + §A architecture refs + §G glossary + §D data model + §X external deps + §R runtime topology + §O observability + §Sec security + §B build/deploy. |\n| `minimal` | Solo project, small surface, low ceremony. Three short files, no tables you'd leave empty. |\n| `strict` | High-stakes codebase. Hard rules, pre-flight checklists, required \"What I did NOT change\" reports. |\n| `karpathy` | The 65-line behavioral spec made viral by Andrej Karpathy's repo. Four numbered principles. Quotable mantras. |\n| `google` | Boring code, small CLs, one-thing-per-change, style-as-contract, design-note-first. |\n\n\n```bash\natlas styles # list available presets\natlas init --style karpathy # use the karpathy preset\natlas init --style strict --force # overwrite with strict mode\n```\n\nMix-and-match works: a style only overrides files it ships. Missing files fall back to the default template.\n\n> Want a deeper preset with a research → critics → gaps → ADR pipeline + tech-stack docs (web / mobile / Rust / Python / ML / CLI / desktop / …)? Fork the repo, drop your variant in `templates/styles//`, and open a PR. See [`docs/CONTRIBUTING.md`](docs/CONTRIBUTING.md).\n\n---\n\n## Conformance — keep the quartet honest\n\n`atlas check` is a real validator, not a linter afterthought. It reports two severities: **errors** (spec MUST violations that break cross-tool reliance — fail, exit 1) and **warnings** (SHOULD / conditional MUST — advisory, exit 0). So a partial setup never blocks you, but CI can still demand the full standard.\n\n```bash\natlas check # human-readable: errors vs warnings\natlas check --json # machine-readable {ok, errors[], warnings[], …} — pipe to any CI/agent\natlas check --strict # promote warnings → errors (the per-commit CI gate)\natlas check --deep # + SCARS anchor-schema: ToC↔body, problem-needs-remedy, Where: paths resolve\natlas fix # auto-resolve what's fixable (kebab the SKILL dir, re-mirror AGENTS.md, regen llms.txt)\n```\n\n`--deep` (opt-in) is the strongest mode: every ToC link has a body and vice-versa, any scar that states a `**Symptom.**` also gives a `**Do.**` remedy, and every `**Where.**` file path resolves on disk (globs and code-block examples are skipped). All `--deep` findings are warnings, so it never breaks an existing repo unless you add `--strict`. Drop `atlas check --strict` (add `--deep` when you want anchor rigor) into a pre-commit hook or CI step and the quartet can't silently rot.\n\n---\n\n## The autonomous loop *(opt-in 5th surface)*\n\nThe quartet is the **static** knowledge — where things live, what breaks, how to act. The optional **5th surface** standardizes the *dynamic* process: how an agent **keeps improving a repo on its own**, iteration after iteration, without re-inventing the rules each time.\n\n```bash\natlas init --loop # scaffold LOOP.md + ROADMAP.md (alongside the quartet)\natlas check --strict # a loop repo still passes — the surface is validated, not just dropped\n```\n\nTwo files get added:\n\n| File | Role |\n|---|---|\n| `LOOP.md` | **The rulebook + one-command entrypoint.** What *one iteration* means, the anti-churn / honesty / red-team rules, and how to run it. |\n| `ROADMAP.md` | **The EV-ranked task queue + a Done log.** Each item carries why · how + exact entry-points · impact · test · complexity. |\n\n**One iteration, in practice:**\n\n```\n1. Read ATLAS §0 + SCARS ToC → orient (no grep)\n2. Pick the top ROADMAP item by EV → edge × P(real) × leverage ÷ cost\n3. Implement the smallest real change → surgical diff\n4. atlas check --strict → conformance gate before commit\n5. Move the item to the Done log; grow SCARS on any new failure mode\n```\n\nThe scaffold bakes in the rules that make a loop *productive instead of churny*: anti-churn pre-flight (grep before building; never rebuild what ships), EV-ranked selection, a novelty mandate, self red-team before commit, measure-then-gate honesty (ship descriptive-only; wire to behavior only after out-of-sample validation), and `atlas check --strict` as the per-commit gate. `atlas check` validates the loop files **only when they exist** (warnings: `ROADMAP_NO_QUEUE`, `ROADMAP_NO_DONE`, `LOOP_NO_ROADMAP`, …) and reports them under `\"loop\"` in `--json`. A repo without a loop is completely unaffected. Full definition: [SPEC §8](docs/SPEC.md).\n\n---\n\n## The AKIGI protocol *(opt-in — how repos talk to each other's agents)*\n\nThe loop makes ONE repo improve itself. The **AKIGI quartet** makes repos improve **each other**: an agent working in a sibling repo can discover what your repo is for, request a capability, disclose a bug, or responsibly report a vulnerability — and reliably learn the outcome. This way the repo *lives*: it improves from ecosystem demand, not just its owner's attention.\n\n```bash\natlas init --intake # scaffold all four: AKIGI.md + FRQ.md + BRD.md + SRD.md\natlas init --frq # or pick surfaces individually (each implies --akigi)\n```\n\n| File | Role |\n|---|---|\n| `AKIGI.md` | **The purpose contract** *(ikigai, for a repository)* — Purpose · Serves whom · Scope · Non-goals · Acceptance principles · Values. ONE document read identically by humans, the repo's own agents, and outside agents. Every incoming request is triaged against it. |\n| `FRQ.md` | **Feature Request Queue** — outside agents file asks (`Requested by · Why · Ask`); the repo's agent replies inline: `✅ RESOLVED` with the exact contract, or `⛔ DECLINED` with the reason + an alternative. |\n| `BRD.md` | **Bugs Responsible Disclosure** — outside agents disclose broken behavior (evidence + repro required); accepted disclosures graduate into the repo's internal `BUGS.md → fix → SCARS` flow. |\n| `SRD.md` | **Security Responsible Disclosure** — the public entry is a minimal marker only; full detail goes through the file's private channel; **never triaged by an agent** — the maintainer is always escalated. |\n\n`atlas check` validates each surface **only when present** (`AKIGI_NO_ACCEPTANCE`, `FRQ_NO_PROTOCOL`, `SRD_NO_CONTACT`, …); the `llms.txt` export lists all four so outside agents can discover them. Generalized from a production cross-repo protocol (one repo's agent filed a request; the target repo's agent shipped the endpoint and replied with the exact contract — then appended a dated ⚠ BREAKING note when auth changed). Full definition: [SPEC §11](docs/SPEC.md). *(Per-agent identity, inboxes, and compensation for verified disclosures are planned protocol phases.)*\n\n---\n\n## Subcommands\n\n```bash\natlas init [--style ] [--force] [--analyze] [--loop]\n # scaffold the quartet (--analyze auto-drafts the map; --loop adds LOOP.md+ROADMAP.md)\natlas check [--json] [--strict] [--deep] # validate the quartet (errors vs warnings); --deep adds SCARS anchor-schema checks\natlas check --changed-files # gate stale-map drift in CI\natlas fix # auto-resolve warnings (kebab the SKILL dir, re-mirror AGENTS.md, regen llms.txt)\natlas measure [--badge] # estimate orientation-token savings (with vs without)\natlas bench [--runtime claude|codex] # A/B a task with vs without ATLAS via a real headless agent\natlas doctor # diagnose install + harness + runtime-export drift\natlas export --to # fan the quartet out: codex|copilot|gemini|cursor|llms-txt|all\natlas badge # print a 'Powered by ATLAS' README badge\natlas anchors # list every SKILL anchor (machine-readable)\natlas anchor add NAME \"summary\" # append a stub anchor\n\natlas install --runtime claude-code # wire ATLAS into an agent runtime\natlas styles # list available --style presets\n\natlas mirror init [--staged|--direct|--dual-repo --public-repo URL]\n # scaffold .atlas/mirror.allow + GH Action\natlas mirror push [--dry-run] # push only allowlisted refs to the `public` remote\natlas mirror status # show config + what would be pushed\n # refuses if remote is named `origin` (origin is private/GitLab)\n\natlas auth login # interactive picker: ssh vs vendor (gh + glab)\natlas auth login --method ssh # generate per-host keys + configure ~/.ssh/config (idempotent)\natlas auth login --method vendor # brew install gh + glab + run their browser auth flows\natlas auth status # show what's authenticated (CLIs, keys, ssh -T tests)\n```\n\nThe CLI also ships subcommands designed for deeper workflows (`atlas critique`, `atlas adr add/list`, `atlas research add/list`, `atlas gap-to-article`, `atlas cost`). These work best with a style that scaffolds the matching files (`docs/adr/`, `research/`, `docs/gaps/`, ATLAS §C / §GPU sections) — see `atlas help` for the full list.\n\n### Mirror push — three patterns\n\n```bash\n# 1) Direct (simplest)\natlas mirror init --direct\n# one GH repo. local main → remote main. trust your allowlist.\n\n# 2) Staged (default — recommended)\natlas mirror init\n# one GH repo. local main → remote `public` branch.\n# GH Action promotes `public` → `main` after gates pass, auto-deletes staging branch.\n# public repo at rest shows only main + tags.\n\n# 3) Dual-repo (maximum paranoia)\natlas mirror init --dual-repo --public-repo git@github.com:you/project.git\n# TWO GH repos. push lands on the PRIVATE staging repo.\n# GH Action there pushes only main + tags to the PUBLIC release repo.\n# public repo NEVER sees branches, PRs, or any other refs.\n# needs a deploy key (PUBLIC_REPO_DEPLOY_KEY secret).\n```\n\n---\n\n## Auth — using GitHub + GitLab side-by-side\n\n`atlas mirror push` is auth-agnostic — it just runs `git push`. You set up auth once for each host. Two clean approaches:\n\n### SSH with per-host keys *(recommended)*\n\n```bash\n# Two separate SSH keys\nssh-keygen -t ed25519 -f ~/.ssh/id_ed25519_github -C \"you@github\"\nssh-keygen -t ed25519 -f ~/.ssh/id_ed25519_gitlab -C \"you@gitlab\"\ncat ~/.ssh/id_ed25519_github.pub # → upload to GitHub\ncat ~/.ssh/id_ed25519_gitlab.pub # → upload to GitLab\n\n# Pin each key to its host\ncat >> ~/.ssh/config <<'EOF'\n\nHost github.com\n HostName github.com\n User git\n IdentityFile ~/.ssh/id_ed25519_github\n IdentitiesOnly yes\n\nHost gitlab.com\n HostName gitlab.com\n User git\n IdentityFile ~/.ssh/id_ed25519_gitlab\n IdentitiesOnly yes\nEOF\n\n# Verify\nssh -T git@github.com\nssh -T git@gitlab.com\n\n# Use\ngit remote add origin git@gitlab.com:you/project.git\ngit remote add public git@github.com:you/project.git\n```\n\n### HTTPS via vendor CLIs *(simpler setup)*\n\n```bash\nbrew install gh glab\ngh auth login # GitHub: browser flow, stores in Keychain\nglab auth login # GitLab: same\ngit remote add origin https://gitlab.com/you/project.git\ngit remote add public https://github.com/you/project.git\n```\n\nPick **SSH** for CI / multi-machine; **HTTPS + gh/glab** for one-machine convenience.\n\nEither way, `atlas auth login` automates the setup — interactive picker, or pass `--method ssh` / `--method vendor` to skip the prompt.\n\n---\n\n## Supported runtimes\n\n\n| Runtime | Status | Adapter |\n| ----------------------------------------------------------------- | ------------- | ---------------------------------------------------------------------------- |\n| [Claude Code](https://claude.ai/code) | ✅ first-class | `[adapters/claude-code/](adapters/claude-code/)` — hook + slash command |\n| [Codex CLI](https://github.com/openai/codex) | ✅ supported | `[adapters/codex/](adapters/codex/)` — `AGENTS.md` global |\n| [OpenCode](https://github.com/sst/opencode) | ✅ supported | `[adapters/opencode/](adapters/opencode/)` — `AGENTS.md` global |\n| [Cursor](https://cursor.com) | ✅ supported | `[adapters/cursor/](adapters/cursor/)` — `.cursor/rules/atlas.mdc` |\n| [Gemini CLI](https://github.com/google-gemini/gemini-cli) | ✅ supported | `[adapters/gemini/](adapters/gemini/)` — `GEMINI.md` global |\n| [Zed](https://zed.dev) | ✅ supported | `[adapters/zed/](adapters/zed/)` — `.zed/atlas.md` + agents.md |\n| [GitHub Copilot Chat](https://github.com/features/copilot) | ✅ supported | `[adapters/copilot/](adapters/copilot/)` — `.github/copilot-instructions.md` |\n| [Hermes](https://github.com/NousResearch/Hermes-Function-Calling) | ✅ supported | `[adapters/hermes/](adapters/hermes/)` — system-prompt fragment |\n| **MCP-native** — OpenClaw · NemoClaw · opencrust · zeroclaw · Qwen · Windsurf | ✅ via MCP | the `atlas mcp` server — one `mcpServers` snippet, [30-sec setup](docs/INTEGRATIONS.md#-the-30-second-way--mcp-works-in-every-modern-agent) |\n| Any other | ✅ generic | `[adapters/generic/](adapters/generic/)` — manual hook |\n\n\nEach adapter is a single idempotent shell script. Run it twice — nothing duplicates. Building one for a new runtime takes about 30 minutes; see `[docs/CONTRIBUTING.md](docs/CONTRIBUTING.md)`.\n\n---\n\n## Why ATLAS\n\nThree pains every agent hits, and what each ATLAS file kills:\n\n\n| Pain | Killed by |\n| ------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------- |\n| **Re-orientation tax.** Every fresh session re-greps the tree, re-reads the same files, re-asks the same questions. | `ATLAS.md` — a one-shot pointer map. |\n| **Forgotten scars.** A bug you fixed last month → next agent makes it again → days lost. | `SKILL.md` — every lesson has a *name* you can cite. |\n| **Drift in conduct.** Each agent guesses your team's norms — coding style, commit format, refactor appetite. | `CLAUDE.md` — the behavioral contract, runtime-agnostic. |\n\n\n---\n\n## Performance — how much does ATLAS actually save?\n\nATLAS is **a harness performance tool disguised as documentation**. Three measurable wins, in order of size:\n\n### 1. Orientation tokens *(every session, easy win)*\n\n\n| Without ATLAS, a fresh agent typically does | Tool calls | Tokens (est.) |\n| --------------------------------------------------------------- | ---------- | ------------------- |\n| `ls -R` / `find . -type f` to map the tree | 1–2 | 5–20k |\n| `git ls-files` then open 3–7 random files to learn the codebase | 4–8 | 30–100k |\n| `grep -r \"\"` to discover an unknown convention | 1–3 | 10–30k |\n| 3–5 clarifying back-and-forth turns with the user | — | 10–30k |\n| **Total orientation overhead per session** | **~10–30** | **~55–180k tokens** |\n\n\n\n| With ATLAS, the SessionStart hook auto-prints | Tool calls | Tokens |\n| -------------------------------------------------------------- | ---------- | --------- |\n| `ATLAS.md §0` (Quick orientation table) + §1 (Top-level files) | **0** | ~1.4k |\n| `SKILL.md` Table of Contents (every anchor + 1-line summary) | **0** | ~400 |\n| **Total auto-loaded — before the first tool call** | **0** | **~1.8k** |\n\n\nThen on-demand `Read` for one specific module the agent needs (1–5k). Net per session: **~5–10k vs ~55–180k.**\n\n> **TL;DR: orientation phase — 12.8× ([measured](docs/benchmarks/RESULTS.md), 92%) on this repo, scaling toward 30× on large codebases.** Total session cost: ~10–25% *(estimate — not yet benchmarked)*. Bigger codebase = bigger win.\n\n### 2. Bugs not re-made *(compounding win, pays dividends over months)*\n\nEvery SCARS anchor (`§NAME-LIKE-THIS`) is a scar another agent will not re-create. **Audit your own value**: `git log --grep \"§\"` — every commit citing an anchor is a turn where an agent (or human) skipped 2–10 minutes of re-debugging a known issue.\n\nA 6-month-old codebase with 20 anchors and an active team can save **5–15 person-hours per week** in not-re-debugging tax. At $150/hr engineering rate, that's $30k–$90k/year per project, just from the SKILL-anchor mechanism.\n\n### 3. Bad PRs not opened *(silent win, human-review-time)*\n\nCLAUDE.md §3 (*\"Surgical changes — touch only what you must\"*) suppresses the **\"while I'm here\" cleanup commits** that bloat PRs and balloon human review time. Estimated from ~50 PRs reviewed across teams adopting an ATLAS-style behavioral contract: **~30–60% fewer non-load-bearing changes per PR** *(estimate — not benchmarked here)*. Not token savings — reviewer-attention savings, which often dominate engineering throughput.\n\n### Where the savings come from *(visualised)*\n\n```\nWITHOUT ATLAS WITH ATLAS\n \n[ user prompt ] [ user prompt ]\n │ │\n ▼ ▼\n ┌──────────────┐ ┌──────────────────────┐\n │ ls -R │ 20k tokens │ SessionStart hook │ 1.8k tokens\n │ git ls-files │ 30k │ ATLAS §0 + §1 + ToC │ 0 tool calls\n │ read x5 │ 50k │ (already in context) │\n │ grep │ 20k └──────────┬───────────┘\n │ clarify x3 │ 30k ▼\n └──────┬───────┘ ┌──────────────────────┐\n │ │ Read one specific │ 1–5k\n ▼ │ module │\n ┌──────────────┐ └──────────┬───────────┘\n │ start work │ ▼\n └──────────────┘ ┌──────────────────────┐\n ~150k tokens │ start work │\n ~20 tool calls └──────────────────────┘\n ~5 minutes ~5–10k tokens\n ~1–2 tool calls\n ~10 seconds\n```\n\n### Honest caveats\n\n- **Per-session fixed overhead of ~1.8k tokens** on every session, including trivial 1-turn ones. For < 5-turn tasks, that's net overhead. For anything bigger, it pays back many times over.\n- **Codebase-size dependent.** Tiny project (< 20 files): ATLAS is light overhead. Medium (50–500 files): 5–15× orientation reduction is typical. Monorepo (5000+ files): the savings dwarf the overhead by 50–100×.\n- **Runtime-dependent.** Claude Code hooks integrate cleanly. Codex / OpenCode / Cursor / Gemini / Zed read AGENTS.md-equivalent at session start. Pure inline Copilot autocomplete cannot be steered (smaller context model — adapter steers Chat only).\n- **No published A/B benchmark yet.** The numbers above are estimates from observed orientation-phase token usage. A real controlled A/B (same task, same agent, with vs without ATLAS) is on the roadmap.\n\n### Use cases where ATLAS pays back fastest\n\n\n| Use case | Why ATLAS wins big |\n| --------------------------------------- | --------------------------------------------------------------------------- |\n| Large monorepo, multiple sub-agents | Each agent gets oriented in 0 tool calls instead of repeating 10–20 |\n| Multi-agent dispatch (lead → workers) | Workers skip the orientation phase entirely; lead pays once |\n| Long-running project, many contributors | SKILL anchors compound; new contributors / new agents learn from past scars |\n| External agent + internal codebase | The agent can't grep a codebase it doesn't know — ATLAS is the bridge |\n| CI / scheduled agents | The agent has no human to ask; ATLAS is its entire orientation |\n\n\n### Use cases where ATLAS is overkill\n\n\n| Use case | Why |\n| ------------------------------------- | ---------------------------------------------------------- |\n| One-file script | The orientation phase is the whole task |\n| Throwaway prototype | Maintenance value is zero; ATLAS overhead doesn't pay back |\n| < 20-file project that rarely changes | Manual `ls` is fine; no compounding |\n\n\n---\n\n## ATLAS is working in your project when\n\n- New agents stop asking *\"where is X?\"* — they read `ATLAS.md §0` instead.\n- Commit messages cite `§ANCHOR-NAMES` instead of restating context.\n- The same bug doesn't appear twice — its anchor exists and gets cited from the next attempt.\n- Diff sizes match request sizes — no surprise refactors.\n- *\"What I did NOT change\"* lines appear in agent reports.\n- Onboarding a new engineer takes hours, not days — they read ATLAS+SKILL and they're oriented.\n\nIf you're hitting four of these six, ATLAS is paying for itself.\n\n---\n\n## Spec\n\nThe formal spec is at `[docs/SPEC.md](docs/SPEC.md)`. The short version:\n\n- `**ATLAS.md` MUST live at repo root** and contain §0 (quick-orientation table), §1 module index, §7 cross-cutting concerns.\n- `**SKILL.md` MUST live at `.agents/skill//SKILL.md`** with a Table of Contents linking to stable `§ANCHOR-NAME` anchors. Each anchor body: Symptom → Root cause → Do NOT → Do → Where → Commit SHA.\n- **Anchors are immutable once published.** Rename = new anchor; old one becomes a redirect.\n- `**CLAUDE.md`** behavior contract lives at repo root with an `**AGENTS.md**` mirror for runtimes that look for that name.\n- **All four files updated in the same commit** as the structural change they describe.\n\n---\n\n## ATLAS vs other documentation patterns\n\n\n| Concern | **ATLAS** | `awesome-claude-prompts` | `CLAUDE.md` only | `ARCHITECTURE.md` only | None |\n| ------------------------------------------------------------ | ---------- | ------------------------ | ---------------- | ---------------------- | ---- |\n| Lives in the repo | ✅ | ❌ | ✅ | ✅ | — |\n| Read by Claude / Codex / OpenCode / Hermes | ✅ all | only Claude | only Claude | inconsistent | none |\n| Three orthogonal axes (structural / procedural / behavioral) | ✅ | ❌ | behavioral only | structural only | — |\n| Stable anchors citable from commits | ✅ `§NAME` | ❌ | ❌ | ❌ | — |\n| Auto-loaded on session start | ✅ via hook | ❌ | depends | depends | — |\n| Style presets (`karpathy`, `google`, `strict`, `minimal`, `default`) | ✅ | ❌ | ❌ | ❌ | — |\n| CLI for ADR / research / critique / cost / mirror | ✅ | ❌ | ❌ | ❌ | — |\n| Cross-runtime (Claude / Codex / OpenCode / Hermes / generic) | ✅ | partial | ❌ | ❌ | — |\n| Zero deps (bash only) | ✅ | ✅ | ✅ | ✅ | — |\n\n\nATLAS doesn't replace your CLAUDE.md — it formalises *where the rest of the docs go* and gives every agent the same Table of Contents.\n\n---\n\n## Examples\n\nSee [`examples/sample-project/`](examples/sample-project/) for a minimal quartet (a hypothetical TypeScript API project — `User vs Account` SKILL anchor, Postgres data model, runtime topology).\n\n## Used by\n\n\n\n- _(your repo here)_\n\n## FAQ\n\n
\nWhy three files instead of one CLAUDE.md?\n\nThe three files cover orthogonal axes:\n\n- **ATLAS.md** — *structural.* \"Where does X live?\"\n- **SKILL.md** — *procedural.* \"What did we learn the hard way?\"\n- **CLAUDE.md** + **AGENTS.md** mirror — *behavioral.* \"How should the agent act?\"\n\nMost projects today have either a giant CLAUDE.md that mixes all three, or only the behavioral file. Mixing the axes makes each one worse: structure is too detailed for a system prompt; scars and rules don't belong in an architecture doc; behavior buried under module listings doesn't get read.\n\nThree small files, three jobs, one cross-link convention.\n\n
\n\n
\nHow is this different from awesome-claude-prompts, llms.txt, AGENTS.md, etc.?\n\n| | ATLAS | awesome-claude-prompts | llms.txt | CLAUDE.md / AGENTS.md only |\n|---|---|---|---|---|\n| Lives in your repo | ✅ | external | external | ✅ |\n| Three orthogonal axes (structural / procedural / behavioral) | ✅ | ❌ | partial | behavioral only |\n| Stable anchors citable from commits | ✅ `§NAME` | ❌ | ❌ | ❌ |\n| Auto-loaded at session start | ✅ via hook | ❌ | varies | depends |\n| Style presets | ✅ 5 | ❌ | ❌ | ❌ |\n| CLI for ADR / research / critique / cost / mirror | ✅ | ❌ | ❌ | ❌ |\n| Cross-runtime (9 adapters) | ✅ | partial | partial | partial |\n| Zero deps (bash only) | ✅ | ✅ | ✅ | ✅ |\n\nATLAS isn't replacing your CLAUDE.md / AGENTS.md — it formalizes *where the rest of the docs go* and gives every agent the same Table of Contents.\n\n
\n\n
\nDoes it work with my agent runtime?\n\nOut of the box: **Claude Code, Codex CLI, OpenCode, Cursor, Gemini CLI, Zed, GitHub Copilot Chat, Hermes**, plus a `generic` adapter for anything that can read a file. Each adapter is a single bash script in `adapters/`. If yours isn't listed, open an issue with the **🔌 New runtime adapter** template — they're usually 30–60 lines.\n\n
\n\n
\nHow does it actually save tokens?\n\nThe SessionStart hook auto-prints `ATLAS.md §0 + §1` (~1.4k tokens) and `SKILL.md`'s Table of Contents (~0.4k tokens) — about **1.8k tokens before the first tool call**.\n\nWithout ATLAS, the agent typically runs `ls`, `find`, `git ls-files`, opens 3–5 files, and asks 3–5 clarifying questions to orient itself — adding up to ~55–180k tokens across 10–30 tool calls.\n\nNet: **12.8× orientation reduction ([measured](docs/benchmarks/RESULTS.md), 92%) on this repo** — scaling toward 30× on large codebases; ~10–25% on total session cost *(estimate)*. See the [Performance section](#performance--how-much-does-atlas-actually-save) for the full breakdown + honest caveats.\n\n
\n\n
\nWhat about agents that aren't allowed to read arbitrary files at startup?\n\nThe hook is one option. The other is: include the ATLAS §0 + SKILL ToC inline in the agent's system prompt. The `adapters/hermes/install.sh` shows the system-prompt-fragment pattern. Every runtime has *some* way to inject project context — ATLAS just gives that injection a stable shape.\n\n
\n\n
\nDoesn't this become outdated as the code changes?\n\nThat's why the spec requires **`ATLAS.md` updated in the same commit as any structural change** (SKILL §ATLAS-IS-INDEX). The `atlas check` command validates the file still parses; the CI integration documented in `docs/INTEGRATIONS.md` fails the build if a structural diff isn't matched by an ATLAS update.\n\nA stale ATLAS is worse than no ATLAS — the spec is opinionated about this on purpose.\n\n
\n\n
\nWhy bash? Why not a \"real\" language?\n\nThree reasons:\n\n1. **Zero install dependencies.** Every system ATLAS targets has bash. No Node-version mismatch, no Python virtualenv, no Rust toolchain. Just `curl | bash`.\n2. **The CLI is small.** ~2000 lines of bash is readable, debuggable, and forkable. A Node/Python rewrite would be 3–5× larger.\n3. **The bash CLI calls out to the right tools when needed.** `git`, `gh`, `glab`, `codex`, `claude`, `jq` — they're already installed on dev machines.\n\nIf you want a Python or Node port, fork it — but the philosophy is \"minimum infrastructure to ship a convention.\"\n\n
\n\n
\nIs this just AGENTS.md with extra steps?\n\nAGENTS.md is one file: behavioral. ATLAS is three files (structural + procedural + behavioral) + a CLI + style presets + runtime adapters + a spec. ATLAS *includes* an AGENTS.md mirror (`CLAUDE.md` is the canonical, `AGENTS.md` is the byte-identical mirror) so Codex / OpenCode / Cursor pick it up natively.\n\nIf you want to start with just `CLAUDE.md` and ignore the rest, fine — the spec is opt-in section by section. But the orientation savings come from §0 + §1 + the SKILL ToC, not from the behavior file alone.\n\n
\n\n---\n\n## Citing ATLAS\n\nIf you use ATLAS in academic work, please cite the preprint (GitHub's\n\"Cite this repository\" button reads [`CITATION.cff`](CITATION.cff)):\n\n```bibtex\n@article{atlas2026,\n title = {The Agent Orientation Tax: Amortizing It with ATLAS,\n a Precomputed Context Layer for Stateless AI Agents},\n author = {Abbasi, Alain},\n year = {2026},\n note = {Preprint. arXiv:XXXX.XXXXX},\n url = {https://github.com/Abbasi-Alain/atlas}\n}\n```\n\n> The arXiv ID / DOI will be filled in once the preprint is live; until then the\n> `note`/`url` point readers to this repository.\n\nTo cite the **software** specifically, use the archived release on Zenodo —\n[DOI 10.5281/zenodo.20766338](https://doi.org/10.5281/zenodo.20766338) (the concept\nDOI always resolves to the latest version):\n\n```bibtex\n@software{atlas_software,\n title = {ATLAS: a precomputed context layer for stateless AI agents},\n author = {Abbasi, Alain},\n year = {2026},\n publisher = {Zenodo},\n doi = {10.5281/zenodo.20766338},\n url = {https://doi.org/10.5281/zenodo.20766338}\n}\n```\n\n## License\n\nMIT — see [LICENSE](LICENSE). Use freely in commercial, OSS, or private agent stacks.\n\n## Contributing\n\nCross-project lessons applicable to most codebases → PR them into `[templates/SKILL.md.tmpl](templates/SKILL.md.tmpl)`. New runtime adapters → drop into `[adapters//](adapters/)`. New style presets → drop into `[templates/styles//](templates/styles/)`. See `[docs/CONTRIBUTING.md](docs/CONTRIBUTING.md)`.\n\n---\n\n## Star history\n\n[Star History Chart](https://star-history.com/#Abbasi-Alain/atlas&Date)\n\n---\n\n*ATLAS is a convention as much as it is code. Spread it.*\n\n**⭐ if you'd use this in your next project. Open an issue if you wouldn't — that's how the spec improves.**\n","readmeFilename":"README.md"}