feat(reflect): make reflection evolution-aware (revise/close desires)

[oratis]

Stacked PR (2/3). Base is claude/desire-web-reflect (#242), not main — GitHub retargets it to main automatically once #242 merges. Merge order: #242 → #246 → #247. The diff below is only this PR's own changes.

Why

Reflection is the path that turns a conversation into a desire — but it could only desire_add, and it was never shown the desires it already had. So it appended near-duplicates blind, and the set only ever grew. Combined with the display always surfacing the same filesystem-first actionable desire, this is a second reason the "wish" felt frozen (root cause #3 in the plan doc).

What

1. Reflection now sees its own desires. A compact ## your current desires block (slug + what + why + actionable) is added to the reflector prompt, so it can target existing ones instead of guessing.
2. Two new operations:
   • desire_revise { slug, what?, why?, actionable?, heartbeat_prompt?, pursuit? } — read-modify-write by slug. Only the supplied fields change; bornAt/slug are identity; an undefined in the patch never wipes an existing value; throws if the slug doesn't exist (no accidental create-via-revise).
   • desire_close { slug, outcome, reflection } — soft close: actionable=false + a [CLOSED:<outcome>] progress note + a [DESIRE_CLOSED] journal line. The file is retained and git-tracked — reversible, nothing destroyed. This is the guardrail that lets reflection prune without violating soul sovereignty.
3. Guidance tells reflection to prefer revising/closing over piling up duplicates.
4. One shared path: the close logic lives in store.closeDesire(), now used by both reflect's desire_close op and the pre-existing desire_close tool (refactored onto it — removes the duplication).

Design rationale (正反方辩论)

The plan doc's Debate 2 weighs append-only (protect sovereignty) vs allow revise/close (enable evolution). Decision: allow, behind four guardrails — reflection is shown existing desires (informed, not blind); close is soft; no hard-delete op exists; every change is git-committed and recoverable. Full debate in docs/PLAN_DESIRE_EVOLUTION_v1.0.md §4 (landed in #242).

Testing

• src/soul/desire-evolve.test.ts — 7 tests: partial revise preserves untouched fields + identity, undefined never wipes, dormant→actionable flip becomes auto-pursuable, soft-close retains the file + writes progress/journal notes, both throw on unknown slug.
• npm test: zero regressions — same 5 pre-existing failures as main, +7 new passing tests. Typecheck clean (pre-existing imapflow optional-dep error only).

Merge order

Independent of #242, but the series reads best merged #242 → this → (PR3).

🤖 Generated with Claude Code