Global, cross-repo agent skills: capture durable session learnings (memory.remember), periodically compact MEMORY.md (memory.compact), run a personal five-phase dev workflow (workflow), run a read-first workspace health check (checkup), and grade candidate models against eval golden sets (evals).
Manual workspace checkup: a read-only health report for the current repo and the user's skill wiring, inspired by a /checkup command. Use when asked to check, audit, clean up, or tune the workspace -- phrases like "checkup", "workspace health", "are my skills in check", "is memory compacted", "AGENTS/PRODUCT/DESIGN…
Runs model exams against the eval golden sets deposited by the workflow skill: "evals.run [candidate model]" exams a candidate on one seat's cases (seats: spec, plan, reviewer, mechanical), grades with the incumbent strict reviewer plus a human spot check, scores quality AND cost/latency, and writes a one-page…
Compact and reconcile active MEMORY.md by grouping entries by Topic, identifying stale, superseded, duplicate, archive-only, and skill-promotion-candidate entries, and producing proposed cleaned files. Reads CE artifact directories and existing skill frontmatter as context. Run manually when MEMORY.md feels bloated or…
Capture durable learnings, new feature behavior, and operating rules from the current session, then write them directly into the current repo's MEMORY.md and related repo-local docs in one pass — including promoting reusable principles into a skill (new or existing) and refining DESIGN.md in place when the repo has…
Trigger ONLY on a message of the form "workflow " or "/workflow " — a deliberate invocation of this personal five-phase solo-dev coding workflow; never on casual mentions of spec, plan, execute, review, or learn elsewhere in a message. Commands: brainstorm, improve, spec, plan, execute, review, learn, wrap, status…