guidelines
Workspace Audit And Reporting
Read-only Agent procedure for workspace health checks, task metadata audits, status reports, and diagnostic summaries.
Purpose
This guideline consolidates the old read-only audit and status-report role. It helps Agents diagnose workspace structure, task quality, content health, and delivery state without silently changing files.
Audits are read-only unless the user has authorized the repair scope through guidelines/proposal-and-dry-run.
Agent Skill
When To Use
Use this skill when the user asks for:
- workspace health, schema, link, or placement diagnosis;
- task metadata, readiness, blocker, owner, or board consistency checks;
- project status, release readiness, delivery progress, or knowledge-base report;
- stale, contradictory, duplicate, orphaned, or localized-only content detection;
- dry-run repair suggestions without applying them.
Bootstrap
Follow the common bootstrap in skills get workspace-operations (guidelines/forma-workspace-operations) using this repository's CLI invocation. Reuse current results and load only the guideline needed for this request.
Run cargo run -q -p forma-cli -- check --json for validation findings. Inspect affected entries; for Task audits also list candidate Tasks and render the board. Reuse unchanged results.
Audit Rules
- Separate machine diagnostics from human judgment.
- Separate field-based, board-based, link-based, git-based, test-based, and inferred evidence.
- Treat task
statusas board membership andreadinessas executability. - Treat proposals as unaccepted until their status and related canonical target show acceptance.
- Treat repository-excluded local files as workflow-local context unless the user selected them for review. Do not claim that Forma paths provide privacy.
- Report stale or contradictory claims with source paths instead of choosing silently.
- Apply repairs only within user-authorized scope, using the preview decision in guidelines/proposal-and-dry-run.
Report Shape
Use the smallest report that answers the question:
Scope: spaces, views, files, or task set inspected.Evidence: commands and key files used.Findings: ordered by impact, with file paths.Dry-run fixes: concrete repairs that would require approval.Requires judgment: decisions the Agent cannot make safely.Residual risk: checks not run or evidence not available.
For status reports, include Reliability: verified, partial, or inferred.
Handoff To Write Work
When an audit finds repair work:
- Summarize the smallest coherent fix.
- Point to the target guideline: guidelines/content-maintenance, guidelines/task-selection, or guidelines/proposal-and-dry-run.
- Continue if the user already authorized the repair scope; otherwise present the concrete repair for approval.
Completion Criteria
Complete the audit only when scope, evidence class, impact-ordered findings, reliability, proposed repairs, judgment boundaries, and checks not run are explicit, with no unapproved file changes.