# Project Status > Inspect, validate, generate, and maintain evidence-backed project readiness. Use for weighted readiness, blockers, active and deferred effort, unknown estimates, provenance checks, one-shot live monitoring, dashboard plan/apply/check workflows, repository attachment for Codex and Claude, or local Project Status plugin packaging and verification. Use the separate RunGlance skill for live agent, workflow, test, usage, lock, HUD, and run-receipt tracking. - Skill: `openly-useful/project-status` (Agent Skill, multi-file: 25 files) - Install (CLI): `npx skillmds@latest add openly-useful/project-status` - Raw SKILL.md: https://api.skillmd.com/api/skills/openly-useful/project-status/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: Openly-Useful (https://skillmd.com/u/openly-useful) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/openly-useful/project-status --- # StatusGlance / Readiness Maintain one conservative source of truth for project readiness. Keep weighted scope, evidence validity, live service health, and external wait separate from RunGlance activity. Calling this skill for status means one bounded read of the canonical manifest. Reuse a known path and inspect only changed evidence; do not repeat repository discovery or launch execution. The compatible plugin/command/storage ID remains `project-status`. ## Cumulative delivery stages For Alpha, Beta, and Live readiness, use the optional versioned `delivery` extension in the same manifest; read [delivery stages](references/delivery.md) before configuring it. Ordered stage scopes cumulatively reuse task IDs and weights. Readiness counts only complete tasks with qualifying current evidence and cleared gates, not activity, implementation estimates, or elapsed time. Report missing scope as unknown; never invent a denominator or add stage percentages. Preserve historical completion when current evidence expires. At requested updates and integrated milestones, show stage readiness, each active lane's known progress, next milestone, remaining task/milestone counts by stage, acceptance gaps with owner/next action, and attention items. Do not create a second tracker or a recurring model poll. Status reporting never authorizes follow-on execution; during an already authorized build, hand gaps to the active delivery loop so ready in-scope follow-ons keep moving. ## Preserve the safety boundary - Default to read-only inspection. `plan`, `check`, `find`, `validate`, `summary`, `verify`, and `monitor once` do not edit project files. - Run `apply`, `init`, or `package build` only when the user explicitly asks for the corresponding write. - Refuse ambiguous tracker, route, output, or attachment conflicts. Never overwrite without the explicit overwrite flag shown by the relevant command. - Treat commit, push, deployment, global installation, scheduling, marketplace registration, submission, and public posting as separate actions. Never infer them from audit, attach, package, or monitor requests. - Never award readiness credit from agent activity, elapsed time, a green endpoint, packaging success, or a fresh browser reload. ## Locate and validate the tracker Use the dependency-free CLI: ```sh node /scripts/status.mjs find node /scripts/status.mjs validate --json node /scripts/status.mjs summary --markdown ``` Prefer `.project-status/manifest.json`. If discovery finds multiple candidates, stop and ask which is canonical. Initialize only when requested: ```sh node /scripts/status.mjs init --project "Project name" ``` Read [the methodology](references/methodology.md) before changing weights, earned credit, evidence, task state, or estimates. The strict v1 manifest uses `route`, `initiative`, `source`, `audit`, `totalWeight`, `phases`, global `evidence`, `gates`, and a dependency-edge array. Treat the bundled schema as authoritative; unknown properties and schema versions are errors. Advance `audit.evidenceAsOf` when the evidence set used for scoring changes, and `audit.verifiedAt` only after a real weighted audit. Validation, rendering, packaging, and browser refreshes do not advance either timestamp. ## Generate or maintain `/status` Read [the dashboard contract](references/dashboard.md), then inspect the host application and run a read-only plan: ```sh node /scripts/dashboard.mjs plan --json ``` The bundled adapter emits a framework-neutral static dashboard. Its recommended output is `public/status` for detected web applications and `status` otherwise. Supply `--output ` when project routing requires another location. Apply only when requested: ```sh node /scripts/dashboard.mjs apply --output public/status node /scripts/dashboard.mjs check --output public/status --json ``` Use `--replace` only after inspecting reported conflicts and receiving authorization to replace those exact generated files. The public projection is allowlist-built: it omits internal/restricted locators, private assertions, internal IDs, and non-public integrity metadata. Integrate the generated static route with the host framework when its routing model does not serve directory indexes automatically; do not claim route readiness until the project's build and route checks pass. ## Verify provenance Read [the provenance guide](references/provenance.md). Plan first, then verify offline evidence: ```sh node /scripts/provenance.mjs plan --json node /scripts/provenance.mjs verify --json ``` Network verification requires explicit `--network`. Localhost requires the additional `--allow-localhost` flag. The verifier never edits the manifest or advances timestamps. Do not expose query secrets, credentials, private paths, or raw private evidence in a report. ## Run one live observation Read [the monitoring guide](references/monitoring.md), then run only a bounded one-shot observation: ```sh node /scripts/monitor.mjs once --json ``` Use `--allow-localhost` only for an intentional local fixture. The command does not cache, record, schedule, retry indefinitely, fix, deploy, or publish. Report its observation time separately from manifest freshness. Manifest freshness describes the timestamped snapshot and never implies that evidence locators or live services were just checked. ## Attach across Codex and Claude Read [the platform guide](references/platforms.md). Inspect before writing: ```sh node /scripts/attach.mjs plan --mode symlink --json node /scripts/attach.mjs apply --mode symlink --json node /scripts/attach.mjs verify --json ``` The canonical repository copy is `.agent-skills/project-status`. Discovery paths are `.agents/skills/project-status` and `.claude/skills/project-status`. Use `--mode copy` where repository-relative symlinks are unreliable. Attachment refuses existing targets and never installs globally. ## Package locally Read [the packaging guide](references/packaging.md). From the source initiative, plan or verify without writing: ```sh node scripts/package-skill.mjs plan node scripts/package-skill.mjs verify ``` Build local archives only when requested: ```sh node scripts/package-skill.mjs build ``` The deterministic builder produces a portable skill ZIP, an OpenAI plugin ZIP, a Claude plugin ZIP, and SHA-256 inventories. It does not install or publish them. ## Handoff Report the exact and rounded readiness; phase denominators; active and deferred effort, unknown estimates, recurring work, and typed gate waits; blockers; changed evidence; provenance limitations; live health and observation time; manifest freshness timestamps; and verified `/status` route. If RunGlance was used, report its task result separately. State which operations wrote files and explicitly say when no commit, deploy, installation, schedule, or publication occurred.