Implement initial PyAutoBrain Feature Agent

[Jammy2211]

Summary

Adds the Feature Agent, the growth function of PyAutoBrain: the reasoning agent that reads feature intent from PyAutoMind and decides how the organism should grow. Implements PyAutoMind/feature/pyautobrain/feature.md.

It reasons; it never edits source. It emits a structured FeatureDecision that the existing development workflow (start_dev → start_library/ship_library, start_workspace/ship_workspace) consumes. Built on the Build Agent template, so it follows the established agents/<name>/ shape (concise AGENTS.md + deterministic entrypoint + a capability audit of the organs it reads).

Mind (PyAutoMind feature/*)  →  Feature Agent  →  start_dev
                                              →  start_library / ship_library
                                              →  start_workspace / ship_workspace
       consults ↘                    ↙ consults
        Health Agent            PyAutoMemory (scientific / architectural context)

Three modes

• specific — plan a named prompt for start_dev.
• selection — choose the best next feature task (not just the first), down-ranking in-flight work read from active.md / planned.md.
• difficulty-constrained — select by --difficulty / --model / --budget / --ambitious / --impact.

What it reasons about

• Difficulty — transparent heuristic (repos affected, prompt size, scientific complexity, architectural risk, test burden, memory-context need) → small | medium | large | too-large, with the factor breakdown in every decision.
• Phasing — direct / split-into-phases (emits _phase_1_design.md … _phase_4_docs.md stubs) / research-first / re-home when mis-filed.
• Workflow — library | workspace | combined decided from the prompt's @RepoName references (per PyAutoMind ROUTING.md); combined enforces library-first ordering so the workspace consumes the library PR's ## API Changes summary.
• Scientific/architectural context — maps the task to the relevant PyAutoMemory sub-wiki (lensing/smbh/cti/methods/galaxies) and cites it, never invents it; degrades gracefully if PyAutoMemory is absent.
• Society of agents — for risky / multi-repo / release-bound work it consults the sibling Health Agent (--check-health) rather than querying Heart directly.

Files

• agents/feature/feature.sh — entrypoint: resolves PyAutoMind/PyAutoMemory checkouts, normalises the subcommand (bare path → specific, nothing → selection), optional health consult.
• agents/feature/_feature.py — stdlib-only, never-writes analysis core: discovery, classification, difficulty, phasing, memory routing, FeatureDecision in human + --json form.
• agents/feature/AGENTS.md + MIND_TAXONOMY.md — concise agent doc and the PyAutoMind/PyAutoMemory surface audit (both 116 lines, under the 200-line skill cap).
• agents/_common.sh — sibling-aware resolve_mind / resolve_memory directory resolvers.
• bin/pyauto-brain + root AGENTS.md — register and document the agent.

Validation

Exercised against the 30 real feature/** prompts in PyAutoMind:

• ✅ identifies feature prompts; selects when none specified (down-ranking in-flight work)
• ✅ respects difficulty constraints (easy / strong-model / impact)
• ✅ recommends phasing + emits phase stubs for too-large tasks
• ✅ references PyAutoMemory sub-wikis where appropriate
• ✅ outputs plans compatible with start_dev / ship_*; combined library+workspace detected with library-first ordering
• ✅ bash -n / py_compile clean; edge cases (unknown task → exit 5, --help, --json, CLI dispatch) verified

🤖 Generated with Claude Code

---

Generated by Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.