Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Project-scoped audit playbook for evaluating structure, documentation, timeline realism, velocity, loose work, duplication, blockers, and strategic gaps while staying in project context.
This is a strategy skill at domain altitude. An audit is fundamentally evaluative — it decides what is healthy, thin, stale, duplicated, or at risk, and ranks findings by severity — so its weight sits in Judgment and the Contract, with the Procedure supplying the ordered read.
Activation
The context type is project and the user asks to audit, review, stress-test, inspect health, find gaps, or look for blockers
You need to assess whether the ideal project structure is being maintained: project -> goals -> milestones/sub-goals -> plans -> linked tasks, with documents supporting the strategy and knowledge layer
You need to evaluate timeline realism, project velocity, bumped work, stale work, loose tasks, duplicate work, or missing documentation
The user asks for a critical read without immediately asking you to change project data
You need evidence-based recommendations before deciding whether any cleanup work is warranted
Judgment
The decision spine for reading a project critically.
A project audit is an analysis workflow, not a separate chat context.
Context snapshots can be intentionally limited. Fetch more before making strong "all clear" or "nothing exists" claims.
The best audits synthesize the whole project, not just a list of defects. Explain the shape of the project and why the issues matter.
Procedure
Keep the chat in project context. Do not switch to a project_audit context type.
Start from the current project scope and reuse the in-context project_id when available.
Prefer read-only analysis first; do not mutate the graph just because you found issues.
Build the structural map: identify the project's major documents, goals, milestones/sub-goals, plans, tasks, risks, and calendar anchors. Use the project overview first when it is enough; fetch the graph and targeted lists when the snapshot is thin.
Audit the ideal project chain: each important goal should have either a clear plan, supporting milestones, linked tasks, or a reason why it is intentionally thin. Each plan should link to the tasks that complete it. Completed task sets should imply plan progress; completed plans should support goal or milestone progress.
Audit milestone structure: check whether milestones are being used as meaningful sub-goals, whether milestone-level plans exist when needed, and whether milestones actually help explain progress toward parent goals.
Audit documentation coverage: find goals, plans, milestones, or risky workstreams that lack strategy docs, research docs, requirements, notes, decision records, or other context that would help the user execute. Also flag stale, unlinked, or orphaned documents when they create confusion.
Audit task fit: identify loose tasks that do not relate back to a goal, milestone, plan, document, or clear project outcome. Separate useful one-off admin tasks from work that should be linked, merged, renamed, or removed.
Audit duplicate and overlapping work: look for tasks, plans, milestones, or docs with similar names, objectives, or descriptions. Treat this as a signal to consolidate or clarify ownership, not as proof of a bug unless the evidence is strong.
Audit timeline realism: compare stated due dates, milestone dates, task volume, blocked work, dependencies, and calendar commitments. Ask whether the proper work can realistically fit in the timeline, and name the biggest schedule compression points.
Audit velocity and plan adherence: look for completed task counts, stale active tasks, overdue tasks, repeated date bumps, blocked work, and whether planned tasks are actually getting done. If historical completion data is thin, say what you can and cannot infer.
Synthesize the audit into a ranked set of findings. For each material finding, include evidence, impact, and the smallest practical correction.
Separate observations from recommendations: first report what is true, then suggest what should change.
If the user asks you to fix issues after the audit, switch from analysis to the appropriate write skill, then use that skill's paired direct tools after resolving exact targets and required fields.
Contract
Return the audit as a structured readout, not a flat defect list:
Executive readout — the shape of the project in 2-3 sentences.
Strongest findings — ranked by severity, each with evidence (the specific entities or data inspected), impact, and the smallest practical correction.
Timeline and velocity assessment — realism of dates vs. task volume, blocked work, and what can and cannot be inferred from available history.
Structural gaps — broken or thin links in the goal → milestone → plan → task chain.
Documentation gaps — goals, plans, or risky workstreams lacking the docs needed to execute.
Prioritized next corrections — kept separate from observations; what to change first.
Stop conditions before replying: every material finding cites visible evidence, not generic project-management advice; confirmed blockers are separated from inferred risks; no "all clear" or "nothing exists" claim is made without having fetched enough data to support it; observations are reported before recommendations.
Good audit output usually has: executive readout, strongest findings, evidence, timeline/velocity assessment, structural gaps, documentation gaps, and prioritized next corrections.
Policy
Do not claim an audit is exhaustive unless you actually fetched enough targeted data to support that claim.
Do not mutate project data by default during an audit.
Do not confuse ideal structure with required structure; some projects legitimately have thin hierarchy.
Distinguish confirmed blockers from inferred risks.
Keep audit findings grounded in visible context or tool results, not generic project-management advice.
Do not force every task into a hierarchy. Some one-off tasks are legitimate, but name them as one-off work and explain whether that helps or hurts project clarity.
Do not treat missing documentation as automatically bad. Explain what decision, strategy, or execution risk the missing document creates.
Do not overstate velocity if completion history, update history, or reschedule history is unavailable.
Related Tools
util.project.overview
onto.project.graph.get
onto.task.list
onto.goal.list
onto.plan.list
onto.milestone.list
onto.risk.list
onto.document.tree.get
cal.event.list
Examples
Audit a project for structural and execution gaps
Load project_audit from project context.
Read the project overview or graph plus key task, plan, milestone, and risk lists when needed.
Trace goals to milestones, plans, tasks, and supporting documents.
Summarize the highest-severity findings with evidence.
Recommend the smallest set of changes that would materially improve project health.
Check whether a project is drifting or stale
Inspect task states, recent updates, unresolved risks, and near-term milestones.
Use calendar reads if upcoming events or work sessions affect the assessment.
Report what looks stale, what is slipping, and what is simply absent from the fetched data.
Find loose work and duplicate effort
Search/list tasks, plans, documents, and milestones by project.
Identify tasks with no clear parent outcome and duplicate-looking records with overlapping names or descriptions.
Recommend whether to link, merge, archive, rename, or leave the work as a deliberate one-off.
Audit documentation against goals
Compare major goals and plans against project documents and document tree entries.
Flag goals that need a strategy, research, brief, requirements, or decision document before execution can improve.
Prioritize documentation gaps that block action or create repeated confusion.
Provenance
internal-default — BuildOS's own project-audit playbook; no external source is cited by this skill. No REMOVED/CORRECTED audit trail exists for it.
