Show the current status of a contract (or all in-flight contracts) right in the CLI — phase, SLA time-left (IST + UTC), readiness, and the next action. Use when the user says "status", "where are we", "what's in flight", "time left", or invokes /status [ ].
Verify a contract using a two-layer approach — a deterministic code validator first, then LLM judgment critics. Produces a consolidated findings file with a computed readiness verdict. Use when the user says "verify contract", "check this PRD", "run critics", or invokes /contract verify .
Dev-side critic for the pickup flow. Surfaces auto-fillable defaults the agent can pull straight from the repo + history (Bucket A in contract-pickup), and code-context dev-decides items where engineering judgment is needed (Bucket B). Different from critic-regression — that one flags risks; this one proposes answers.…
Critic that ensures every behavior specifies empty, loading, success, and error states (UI copy + notifications). Invoked by contract-verify. Catches the "we forgot the error message" bug class.
Critic that scans a Failsafe contract for missing edge cases, ambiguous scenarios, and unspecified failure modes. Invoked by the contract-verify orchestrator. Returns findings as a JSON array.
Critic that ensures every behavior carries an analytics event and that success metrics map to a fireable event. The keystone for post-launch verification. Invoked by contract-verify.
Critic that pushes back on disproportionate scope — the counterweight to critics that only ever ADD. Flags a contract that handles far more than the change requires (new flags, analytics taxonomies, shadow-observation baselines, exhaustive speculative edge cases). Invoked by contract-verify; runs ALWAYS, and is the…
Critic that checks every behavior is specified for every platform in scope. Flags web-only specs for cross-platform features. Invoked by contract-verify.
Critic that scans across multiple repos (Flutter, Web, Backend, etc.) to find existing behaviors related to the new contract and flags potential regressions, conflicts, or coordination requirements. Invoked by contract-verify. Reads code via the GitHub MCP (preferred — always current main) or local clones (fallback…
Implement a promoted contract — write code in the target repo's stack, generate tests in that stack's idiom, iterate until they pass, open the PR. Also runs in fix-mode (review→fix loop) to address open PR-review/verify-pr findings on an existing PR. Stack-agnostic: works for web (React/Next/Vue/...), mobile…
Produce a post-launch verdict (landed / partial / not-landed) for a contract by reading telemetry, errors, and user feedback. Invoked manually via /launch or by launch-monitor.yml on cron at day 1/7/14/28. The loop-closer — the thing that answers "did this actually work.".
Express lane for trivial bugs and tiny changes — skips the full contract ceremony. Use for one-line fixes, copy/CSS tweaks, config changes, dependency bumps, or any change localized to one root cause with no new behavior, instrumentation, schema, or platform. Triggered by /fix . Triages size first and escalates to…
The PM-side readiness gate — "airport security for PRDs." A PM (or anyone) gives a PRD as any source — JIRA / Linear / Notion / Confluence / Google Doc / Slack link / GitHub issue / Figma URL, pasted text, an image, or a plain description — and the agent scores it against the Definition of Ready, surfaces the blockers…
During a feature's rollout window, watch Sentry errors, crash-free rate, and release adoption against the contract's budgets and RECOMMEND proceed / hold / recommend-rollback. Never executes a rollback — it proposes and a human approves. Invoked by /rollback-check or the rollback-guard.yml workflow (tight cadence…
Close the loop after a feature is rolled back. Records landed:rolled-back, writes a blameless postmortem from the contract's own timeline + guard reports + Sentry, posts it to Slack, and reopens the work as a follow-up. Invoked by /postmortem after a human confirms the rollback. The one ending the end-to-end loop was…
Check whether the user's environment has the MCPs, secrets, and config Manifest needs. Reports what's connected, what's missing, and what each missing piece would unlock. Invoked by /manifest setup, /setup, on first run, or whenever a skill detects a missing prerequisite. Tells the user what they can do today and what…
Interactive first-run setup. Auto-detects the user's repos, frameworks, event SDKs, and test patterns, asks ONLY the questions detection can't answer, then writes a complete .manifest/repos.yml with zero placeholders. Use on first run, when /setup finds no repos.yml, or when the user says "set up manifest", "configure…
Interactive 3-minute walkthrough of Manifest for first-time users. Use when a user types /manifest, /manifest tutorial, /contract (with no args), /contract help, asks "how do I use this", or otherwise signals they're new and want orientation. Adapts depth based on what the user wants to learn.
Verify that a deployed environment actually delivers what the contract promised. Routes per-platform — Playwright against a URL for web; Flutter integration tests; native iOS (xcodebuild) / Android (gradlew) with Firebase Test Lab; API smoke tests for backend — plus telemetry sampling (Firebase/Crashlytics/Sentry).…