deep-work

Nightly offline self-evolution for this repo's Claude agent: harvests past Claude Code sessions (read-only), mines recurring tasks, replays them offline on your own API budget, and consolidates what it learns into validated CLAUDE.md memory and SKILL.md skills behind a held-out validation gate, staged for review (nothing live changes without /skillopt-sleep adopt). Default backend is mock (no API spend); vendored from microsoft/SkillOpt with safety-hardening patches -- see attribution.derivation_note. Use when the user wants their agent to learn from how this repo's skills are actually used, asks about a nightly/offline sleep or dream cycle, memory/skill consolidation, or invokes /skillopt-sleep.

Convert a rambling description of a desired outcome into one polished, autonomous /goal prompt ready to paste into a fresh session. Use when the user says '/fable-goal', 'turn this into a goal prompt', 'write me a fable prompt', or rambles about something they want made and asks for the prompt that makes it happen. Extracts deliverable/quantity/stakes/tools/destination, asks at most one question batch, verifies every named resource against the live environment, writes a 150-350 word prose prompt with the seven-part anatomy (desire + stakes, quality bar, tool inventory + discovery mandate, creative-freedom grant, medium-matched verification loop, delivery, goal line + autonomy directive), and self-checks six binary criteria before delivering. The output is a single copy-paste prompt, never the build itself.

Personal meeting hygiene, end to end. Before the meeting: a cost gate that prices the meeting in real dollars (attendees x minutes x rate, optionally + the 23-minute refocus overhead per attendee) and refuses to let it exist without a decision, an agenda, and a named owner — verdicts are ASYNC / NOT-READY / MEET, never a shrug. Then a timeboxed agenda builder that rejects any topic without a desired outcome, orders decision topics first, enforces a 5-minute closing actions-recap buffer, and refuses overflowing agendas by name. After the meeting: a deterministic action-item extractor that turns raw notes into an owner + due-date checklist and flags ORPHAN (no owner) and NO-DUE items, because an action item without an owner and a date is theater. Built on the Rogelberg / HBR / Grove meeting-science canon with stdlib-only scripts.

Run a GTD-style weekly review that actually closes loops. Walks David Allen's three-phase ritual — GET CLEAR (collect, process inboxes to zero, empty your head), GET CURRENT (next actions, previous + upcoming calendar, waiting-for, project lists), GET CREATIVE (someday/maybe, new ideas) — backed by three deterministic stdlib scripts: an open-loop scanner that inventories unchecked checkboxes, TODO/FIXME markers, and stale files; a checklist gate that computes completion, names every missing step, and refuses a COMPLETE verdict while any GET CURRENT step is unaccounted for; and a commitment auditor that flags STALLED, NO-NEXT-ACTION, and SOMEDAY-CANDIDATE projects with a 0-100 commitment-health score. The periodic-review counterpart to capture (intake) and reflect (per-conversation reflection).

Self-Improving Agent: curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills. Provides /si:memory-review, /si:promote, /si:extract, /si:memory-status, and /si:remember slash commands.