Questions that come back

Empty and not-run look identical in a total. The detector list above the findings, and the status on each pipeline step, are where the difference is drawn — a pillar with no successful detector reads "not checked", and a count from an incomplete run is labelled a floor.

If a snapshot hit an ingestion ceiling, that is banner-ed above the counts and every rule that would have passed reports UNKNOWN instead.

• No model was reachable for this run, so the reasoning steps produced schema-valid output with every field unknown rather than inventing one.
• A claim’s citations did not resolve against the snapshot, so it was demoted by the evidence gate.
• The snapshot was truncated, so rules that need completeness declined to conclude.

This installation reads snapshots. It does not install dependencies, build, run a test suite or execute a script from a manifest — so coverage metrics and script-based constraints are reported as NOT_SUPPORTED with the reason, rather than estimated.

That also means nothing here writes to your repository. The CLI writes files on your machine, when you run it.

Every Skill carries a separate review with typed issues, and generic language is one of them. pskl skills show <slug> prints the review above the document on purpose: it is the reason to trust or distrust what follows.

A Skill written from a truncated or model-less run has less to be specific about. Re-running the analysis with a provider connected is usually the fix; regenerating without one is not.

The file keeps loading after the code moves, nothing errors, and the agent answers confidently about a shape the project no longer has. `pskl skills freshness` names the files each Skill draws on that have changed, and pskl keel stale does the equivalent for a design session.

• Automatic synchronisation with a repository — an analysis happens when you ask for one.
• Pull-request intelligence and commit analysis.
• Autonomous coding, or any write path into your repository — including from the Review Board.
• A Skill marketplace, team accounts, and editor extensions beyond the read-only MCP connection.

These are different from the four things it does not do at all, which are not on any roadmap. Where a capability is planned but absent, the interface renders it as not built rather than as pending work — the same rule the pipeline follows for a step the worker is never going to run.