feat: add private memory hygiene discipline

[ruby-dlee]

Intent

Build Firstmate's memory-hygiene discipline so private data/captain.md and data/learnings.md stay lean and robust. Add a tight agent-only internal skill, roughly 60-80 lines, that owns per-entry rule-or-fact-only essence, strips emotional or dramatic incident narrative, permits at most a one-line example or pointer, deduplicates into one owner, and applies inspect-then-update with in-place pruning; include a concise before/after example. Sharpen only AGENTS.md section 6's two private-memory routing rows plus a one-line pointer, and add the matching section 13 load trigger. Keep routing, stow behavior, and repository-wide one-owner guidance with their existing owners, do not duplicate the full standard into AGENTS.md, do not touch the actual gitignored private memory files, and keep the diff single-concern. Ship a PR but do not merge it.

What Changed

• Add an agent-only memory hygiene skill that keeps private memory entries concise, deduplicated, action-oriented, and free of incident drama.
• Update knowledge-routing guidance and skill triggers for changes to data/captain.md and data/learnings.md.

Risk Assessment

✅ Low: The change is narrowly scoped and satisfies the source-verifiable intent: it adds a 67-line internal memory-hygiene skill, preserves existing ownership boundaries, updates only the permitted AGENTS.md locations, and does not touch private memory files.

Testing

The already-green full baseline was supplemented with the focused stow regression and an end-to-end trigger-to-curated-memory acceptance check; the 67-line internal skill, ownership boundaries, single-concern scope, and untouched private-memory files all passed, with Markdown evidence captured instead of a screenshot because this is a non-UI agent-instruction change.

Evidence: Memory-hygiene end-to-end evidence

# Memory-hygiene end-to-end evidence

Target: `5ea5640d9f03e660cf6ff87d8c7c63246a4cd015`

This artifact follows the actual operator path introduced by the change: the private-memory routing row points to the internal skill, section 13 declares the matching load trigger, and the loaded skill shows the final curated memory shape.

## Trigger surface seen by Firstmate

From `AGENTS.md` section 6:

> Load `memory-hygiene` before writing or leaning `data/captain.md` or `data/learnings.md`; it owns the per-entry leanness standard while this section owns routing.

From `AGENTS.md` section 13:

> `memory-hygiene` - load before writing, rewriting, pruning, deduplicating, or otherwise leaning `data/captain.md` or `data/learnings.md`.

## Curated result shown by the loaded skill

Input incident narrative:

> 2026-07-22: The captain was furious after an agent tried to deploy while checks were failing, and stated very firmly that this was completely unacceptable because the incident wasted hours and created a stressful rollback.

Resulting private-memory entry:

> Require green CI before deployment. `[[deploy-incident-2026-07-22]]`

The result is one actionable rule with one evidence pointer. It removes emotion, dramatic chronology, and commentary, matching the rule-or-fact-only standard.

## Executed acceptance output

`` `text
ok - stow skill task-note contract includes recoverable body archival
ok - AGENTS.md task-note contract includes recoverable body archival
skill_lines=67
changed_files=.agents/skills/memory-hygiene/SKILL.md,AGENTS.md
private_memory_changes=none
stow_owner=unchanged
one_owner_guidance=unchanged
trigger_chain=section-6-pointer,section-13-trigger,internal-skill
curated_example=Require green CI before deployment. [[deploy-incident-2026-07-22]]
`` `

The acceptance command inspected the committed target, verified the 60-80-line bound, internal/non-invocable metadata, rule/fact-only and anti-narrative constraints, one-line example-or-pointer limit, deduplication, in-place pruning, the section 6 pointer, the section 13 trigger, unchanged stow and repository-wide one-owner owners, no private-memory-file changes, and the two-file single-concern scope.

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

✅ **Test** - passed

✅ No issues found.

• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; uv run --directory tools/agent-fleet --locked pytest || rc=1; uv run --directory tools/agent-fleet --locked python -m compileall -q src || rc=1; exit "$rc"
• Harness-provided successful baseline: command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; uv run --directory tools/agent-fleet --locked pytest || rc=1; uv run --directory tools/agent-fleet --locked python -m compileall -q src || rc=1; exit "$rc"
• bash tests/fm-stow-contract.test.sh
• Commit-level acceptance shell check for skill length/metadata, rule-or-fact essence, anti-narrative guidance, one-line example-or-pointer limit, deduplication, in-place pruning, section 6/13 triggers, unchanged stow and one-owner guidance, two-file scope, and absence of private-memory changes
• Manual review of bf7c5ad167514cba879cb095c091fe1627549ccc..5ea5640d9f03e660cf6ff87d8c7c63246a4cd015 and the rendered trigger → skill → curated-result evidence
• git status --short after testing

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.