Run an admitted change through staged execution where each stage produces required evidence before the next begins. Use after a shape or Formula is accepted and before a diff is offered for review, when one agent would otherwise implement, clean, test, and judge its own work in a single context. Separate…
Test whether behavioral coverage can reject realistic wrong implementations. Use when tests are green but may be weak, after fixing a bug, and whenever code implements or changes a protocol, parser, serializer, state machine, lifecycle, event translation, browser interaction, adapter, or compatibility layer. Also use…
Candidate: Keep a decision trail (a.k.a. show-your-work) while implementing, so a later reviewer can audit the choices and not just the diff. A TSV appended one row per real decision (what, why, evidence), written in the moment, never reconstructed. Use at the start of any implementation task an agent runs unattended…
Assess one already-selected GitHub issue, pull request, or manual request before any factory workflow is admitted. Classify the real intent, preserve evidence and uncertainty, recommend the minimum workflow, and stop for human confirmation. Use when a source item may be a question, investigation, unverified bug…
Reassess one or more ongoing engineering workstreams by treating handoffs as leads and revalidating their drift-prone claims against live GitHub state, local Git checkouts, releases, deployments, and package registries. Use when work may have stalled or changed since the last handoff; when the user asks to reassess…
Snapshot the reference graph of a GitHub PR or issue and trace every linked PR, cross-referenced issue, and mention across repos. Use before working an issue/PR to see its whole graph first (the review-full-pr-issue-graph rule); to find orphan, superseded, competing, or duplicate PRs; to see who mentioned or linked…
Survey an unfamiliar or shared issue backlog, qualify a small set of worthwhile contributor-side candidates, compare them in an evidence-backed matrix, recommend one, let the user make the final selection, and seed the canonical Issue Contract. Use for requests like "what should I work on here", "find a good issue to…
Record completed or interrupted maintenance work as an evidence ledger and classify its compiled-knowledge disposition. Use after resolving, disproving, reviewing, or investigating an issue, PR, incident, or agent run; when backfilling an old session; or when another skill needs durable evidence before extracting or…
Gate a diff before pushing or requesting review: deterministic checks first, then focused lenses selected by what the diff changes. Use before opening or updating a PR, when asked to pre-review a change, or after an external review round to harvest each finding into a new gate.
Orchestrate Shaping to decide whether a fix or contributor PR is the right thing to build. Use before implementation when a defect has multiple solution shapes, changes a contract or persistent state, follows a prior review finding, or requires adopting, amending, absorbing, recreating, or rejecting an existing patch.…
Gate whether a fix or existing contributor PR is the right thing to build. Use before implementation when a defect has multiple solution shapes or changes a contract, especially session state, long-lived process configuration, restart or reuse behavior, fixes to prior review findings, and decisions to adopt, absorb…
Gate whether a fix or existing contributor PR is the right thing to build. Use before implementation when a defect has multiple solution shapes or changes a contract, for fixes to prior review findings, and before adopting, absorbing, recreating, or substantially amending an existing PR or patch. In candidate-audit…