SkillsReview & Run

Agent Skills

Review & Run

Browse, review, and trigger skill mining runs from the Skills page on SpecStory Cloud.

The Skills page on cloud.specstory.com is where you review candidate skills, approve the ones worth keeping, and trigger the background runs that mine your sessions.

Open the Skills page

Sign in to cloud.specstory.com and click Skills in the left nav. The page has two tabs: Library and Run Activity.

The Library tab

A single list of every skill across its lifecycle — Review, Ready, and Installed — each with a colored badge. Filter by state with the chips at the top (All / Review / Ready / Installed), and search by name or trigger.

An overview strip shows N need review · N ready · N installed so you can see at a glance what needs your attention.

Review skills (Keep / Dismiss)

Skills in the Review state are borderline candidates the miner wants a human to confirm. Each row has:

  • Keep — approves the skill. It gets forged into your library as a Ready skill you can install.
  • Dismiss — rejects it. It won't appear again.

Click a row to expand it and see:

  • Why you can trust it — confidence score, "judged cross-vendor", session-evidence count, and (for theme skills) prevalence and outcome lift.
  • What it does — the full SKILL.md content.
  • Provenance — the cluster key it was mined from.

Review the evidence, then Keep or Dismiss.

Ready & Installed skills (Pull)

For skills in the Ready or Installed state, the row has a Pull button. Clicking it copies a command to your clipboard:

specstory skills pull <skill-name>

Run that command in a terminal on the machine where you want the skill installed. The web page can't write to your disk, so it hands you the CLI command. See Installation for what that command does.

The Run Activity tab

This tab shows the background mining runs that produce your skills. Each run card shows:

  • Status — queued → sharding → judging → reducing → done (or failed).
  • Live timing, sessions mined, shard progress, and dossier count.
  • Verdict breakdown — how many candidate skills were confirmed, needs-edits, refuted, or unverified.

Expand a run to see the skills it produced (with per-skill verdict and confidence) and a per-shard breakdown.

Trigger a run

Click Run now to start a new mining run. The button is disabled while a run is already in progress. While any run is in flight, the panel auto-refreshes every few seconds and shows a ● live indicator.

Note

Runs are server-side and operator-gated. You don't need the CLI running for a run to proceed — once triggered, it runs in the cloud over your synced sessions.

How skills get to your library

sync sessions  →  mining run  →  candidate dossiers  →  judge  →  Review (you decide)
                                                              →  confirmed  →  Ready (install)
                                                              →  refuted     →  dropped
  • Confirmed dossiers become Ready skills automatically.
  • Borderline (needs-edits) dossiers become Review skills for you to Keep or Dismiss.
  • Refuted or unverified dossiers are run diagnostics — they never become rows.

The more sessions you sync, the more material the miner has to work with.