fix(learn): teach /learn the full CONTRIBUTING.md skill standards

[teknium1]

Summary

/learn now teaches the full set of CONTRIBUTING.md skill-authoring rules — and makes the model actually enforce the ≤60-char description instead of just stating it.

Root cause (#52367): the /learn prompt taught a subset of the HARDLINE standards and mentioned "description ≤60" without any enforcement, so the model wrote full sentences (up to 202 chars across 4 reported runs). The 60-char system-prompt skill index then silently truncates those, dropping the routing signal past char 60.

Changes

• agent/learn_prompt.py — broadened _AUTHORING_STANDARDS:
  • description: added the index-truncation rationale, a count-and-trim self-check ("after you write it, COUNT the chars; if over 60, cut it before saving"), and a good/bad length example. This is the lever for the reported overshoot.
  • platforms gating — declare platforms: for OS-bound primitives (osascript/apt → OS; /proc, os.setsid, SIGKILL → linux; fcntl/termios → POSIX), prefer cross-platform first.
  • author credits the human first, "Hermes Agent" second.
  • rounded out the Hermes-tool framing with the full wrapped-tool mapping (read_file/search_files/patch/write_file/web_extract…) and references//templates/ layout.
• tests/agent/test_learn_prompt.py — new contract test locking in coverage of the full standard set (description self-check, platforms, author, tool framing, scripts layout).

Validation

	Before	After
Rules taught	description (no enforcement), section order, partial tool framing	+ count-and-trim self-check, good/bad example, index rationale, platforms gating, author-first, full tool mapping
Tests	11	12 — tests/agent/test_learn_prompt.py all pass

Prompt-only fix, in keeping with the /learn design (no engine, no model-tool footprint — the agent authors via skill_manage).

Closes #52367.

Infographic

[github-actions]

🔎 Lint report: hermes/hermes-5f8c20c4 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11268 on HEAD, 11268 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5942 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[tonydwb]

Code Review Summary

Verdict: Approved

Prompt-only fix that broadens the /learn authoring standards to cover the full CONTRIBUTING.md ruleset. Well-structured and thoroughly tested.

Looks Good

• Description enforcement: The count-and-trim self-check and good/bad examples directly address the reported overshoot (descriptions up to 202 chars)
• Platform gating: New guidance on declaring platforms: for OS-bound primitives with clear examples (osascript/apt, /proc, fcntl/termios)
• Author credits: Clarifies human-first attribution
• Tool framing: Expanded the wrapped-tool mapping to cover all Hermes tools, not just the basics
• Test coverage: New test_teaches_the_full_hardline_standards locks in coverage of all key rules
• Prompt-only change with no engine or model-tool footprint — aligns with the /learn design

---

Reviewed by Hermes Agent