claude-automation-recommender
==============================

This skill is vendored from the `claude-code-setup` plugin published by
Anthropic and is redistributed here under the terms of the Apache License,
Version 2.0 (see LICENSE in this directory).

Upstream:
  Plugin name:    claude-code-setup
  Upstream version: 1.0.0
  Author:         Anthropic <support@anthropic.com>
  Vendored into:  .claude/skills/claude-automation-recommender/
  Vendored on:    2026-04-25

Modifications:
  See "Local changes" below.

Local changes:
  - 2026-04-25: Added "Working with the baseline" section to SKILL.md.
    Lists the 14 hooks / 8 subagents / 26 skills / 4 commands / 2 MCPs the
    baseline already provides, instructing the recommender to surface
    *gaps* and project-specific tailoring rather than duplicate baseline
    components. Also added a JSON output schema for /init-project to
    consume the recommender's findings programmatically.
  - 2026-04-28: Refreshed "Working with the baseline" counts in SKILL.md
    to match the current shape: 14 → 17 hooks (the 3 lifecycle hooks
    `memory_session_start`, `memory_stop`, `memory_pre_compact` were
    added between 2026-04-25 and now), 34 → 36 skills (the `chore`
    alternate track and `memory-flush` memory-curation skill were added).
    The historical 2026-04-25 entry above is preserved verbatim as a
    record of the snapshot at that date.
  - 2026-04-28 (later): 36 → 37 skills — the user added a vendored
    `developer-audience-context` skill that `technical-tutorials` now
    references as its audience-loader. Updated SKILL.md skill count and
    breakdown; added the new mandatory-sub-skill chain
    (`technical-tutorials` → `developer-audience-context`) to the
    bullet's "Several skills mandatorily invoke another skill" list.
  - 2026-04-28 (final): 37 → 36 skills — consolidated the standalone
    `developer-audience-context` skill into `technical-tutorials` per the
    user's call ("not generic enough; exists solely for technical-tutorials").
    Audience-context content moved into `technical-tutorials/references/`
    as `audience-context.md`, `audience-template.md`, `audience-example.md`,
    and `audience-context-README.md`. Apache 2.0 attribution preserved via
    the consolidated README in that references directory. Updated SKILL.md
    skill count + breakdown; removed the mandatory-sub-skill chain
    (`technical-tutorials` no longer invokes a separate skill — the
    audience reference is local).
  - 2026-05-12: Refreshed the "Working with the baseline" hook count in
    SKILL.md to match the current shape: 17 → 22 hooks. The breakdown is
    now "17 write/run-boundary guards plus 4 lifecycle hooks plus 1
    input-boundary hook". The list of write/run-boundary guards was
    completed (previously omitted `swarm_approval_guard`,
    `spec_design_calls_guard`, and `process_lifecycle_guard`); the
    lifecycle list now includes `harness_continuation` (the workflow
    auto-continuation Stop hook added by the `harness-auto-continuation`
    workflow at `docs/archive/2026-05-12/harness-auto-continuation/`);
    `consent_gate_grant` is named as the input-boundary hook. Skill count
    unchanged at 36. No upstream behavior was modified; only the local
    baseline-inventory paragraph was updated.
  - References/ directory unchanged from upstream.

License:
  Apache License 2.0. Full text in ./LICENSE.

Why vendored:
  The recommender is a mandatory first step in this baseline (see
  docs/init/seed.md §0). Vendoring removes the runtime dependency on
  having the `claude-code-setup` plugin installed and ensures the
  capability travels with the repository.
