mike-arbuzov365
UserA quality-first control plane for governed AI-agent software development.
Categories
Indexed Skills (11)
knowledge-curator
Aggregate existing knowledge-base signals (links, orphans, freshness, schema, privacy, deferred closeout items) into one ranked maintenance backlog with a stable findings ledger, severity, routing and an owner path per finding. Read-only by default. Never merges, never edits source outside the instance it was pointed at.
plan-execution-packet
Plan an execution packet for work too large for one session - an idea, a PRD, an epic, a large refactor, review findings, or cross-project alignment. Produces charter/facts/decisions/roadmap/self-review/start files ready for autonomous execution. Does not implement product code.
plan-idea
Turn a raw request, observation or opportunity into an evidence-based idea artifact with a clear problem, user, minimum useful scope, validation plan, approval source and route. Use before PRD or execution planning when the product idea is not yet durably defined.
plan-prd
Turn an approved idea into an evidence-based Product Requirements Document with users, flows, scope, observable functional acceptance criteria, non-functional requirements, traceability, rollout and approval. Use after idea approval and before non-trivial implementation planning.
knowledge-ingest
Turn raw evidence from work, review, an incident, or a docs audit into a durable knowledge artifact - not a chat summary or raw reasoning. Use after a Knowledge Delta records something durable, or after an incident with prevention value.
knowledge-lint
Manually review knowledge-artifact quality before proposing promotion beyond this project instance, or after a significant packet closeout. Checks source/evidence discipline, type correctness, status/confidence, duplication, links, and privacy.
knowledge-search
Search existing project knowledge before non-trivial work, so a plausible-looking approach doesn't repeat a mistake a prior session already found and recorded. Use as a preflight before implementing, or while planning, to find relevant prior lessons/incidents.
review-execution-packet
Independently audit a packet claimed as executed or closed - verify the claimed integration, evidence, and closeout against fresh repository/tool state rather than trusting the closeout narrative. Use for a post-hoc audit, not while executing the packet yourself.
run-bounded-evidence-loop
Run an iterative engineering task with explicit success evidence, evaluation, iteration and remote-run budgets, adaptation, and stop rules. Use for test/fix/refine cycles; not for unbounded autonomous repetition.
run-execution-packet
Execute an already-planned execution packet's sessions in strict sequential order, as one agent, from its start file through closeout. Use when handed a prepared packet to run - not for planning one.
run-retro
Run a retrospective across many sessions to find patterns that repeat, after a packet closeout, a significant incident, or on a 2-4 week cadence for an active project. Routes each recurring pattern to a knowledge artifact and an explicit promotion decision. Not a second session closeout.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.