Claude Skills by cyb213

Drift audit for tracking files, specs, plans, or citations against live state. Produces one Reviews/ doc with mechanically-verified findings, two-pass (inventory then apply).

Re-explain your own previous answer in plain language, for when it didn't land. Use when the user types /bro, or says the last answer was too dense, too jargon-heavy, or asks for it simpler. Re-expresses what was already said; never answers something new.

Periodic sweep of the workspace's curated narrative (session-note titles + git subjects + roadmap/DECISIONS) to surface repeated, formalizable work and propose turning it into skills or scripts. Writes one Reviews/ doc with a verdict per candidate. NOT correction triage (that's /review-corrections) and NOT drift detection (that's /audit) — this finds un-formalized recurring WORK.

Shape a raw idea into a clear, INTENT-ready brief by interviewing the user one question at a time — reading any existing code/systems the idea touches as ground truth before asking — so it's ready to graduate via /new-project. Use when the user has a new idea to think through ("let's brainstorm X", "kick off an idea", "help me shape X", "I have an idea for…"), or /kickoff.

Run a structured planning session — interview the sources of truth, draft, adversarial review (up to 3 reviewers, scaled to the change), approve, update docs. Required for any change to the shipped engine, a skill/template, a hook, recall, security, or a public release, and for features spanning several files or systems; a small change touching none of those doesn't need it.

Graduate an idea from your lab into its own full-rig project workspace under ~/Projects/<slug> (stamped from the lab template — recall engine, hooks, git drift-gate, vendored identity, seeded memory, genome skeletons). Use when the user says "graduate X", "spin up a project for X", or "/new-project".

Triage pending correction candidates (captured by the recall Stop hook) into memory — promote the worthy, drop the noise, surface the contested. Closes the correction→memory loop.

First-run personalization for a freshly cloned Lab — interview the user one question at a time and write their IDENTITY.md, so the agent knows who it's working with. Use on first run, when IDENTITY.md still has placeholders, or when the user says "/setup", "set up my lab", or "personalize this".

Cleanly wrap a work session — ground state against git+filesystem, update and de-bloat the tracking files, ensure the next-session landing pad is grounded (not invented), scan staged changes for leaked secrets, then commit (and push if there's a remote). Invoke at session end, or whenever the user says "wrap".

Drift audit for tracking files, specs, plans, or citations against live state. Produces one Reviews/ doc with mechanically-verified findings, two-pass (inventory then apply).

Re-explain your own previous answer in plain language, for when it didn't land. Use when the user types /bro, or says the last answer was too dense, too jargon-heavy, or asks for it simpler. Re-expresses what was already said; never answers something new.

Periodic sweep of the workspace's curated narrative (session-note titles + git subjects + roadmap/DECISIONS) to surface repeated, formalizable work and propose turning it into skills or scripts. Writes one Reviews/ doc with a verdict per candidate. NOT correction triage (that's /review-corrections) and NOT drift detection (that's /audit) — this finds un-formalized recurring WORK.

Shape a raw idea into a clear, INTENT-ready brief by interviewing the user one question at a time — reading any existing code/systems the idea touches as ground truth before asking — so it's ready to graduate via /new-project. Use when the user has a new idea to think through ("let's brainstorm X", "kick off an idea", "help me shape X", "I have an idea for…"), or /kickoff.

Run a structured planning session — interview the sources of truth, draft, adversarial review (up to 3 reviewers, scaled to the change), approve, update docs. Required for any change to the shipped engine, a skill/template, a hook, recall, security, or a public release, and for features spanning several files or systems; a small change touching none of those doesn't need it.

Graduate an idea from your lab into its own full-rig project workspace under ~/Projects/<slug> (stamped from the lab template — recall engine, hooks, git drift-gate, vendored identity, seeded memory, genome skeletons). Use when the user says "graduate X", "spin up a project for X", or "/new-project".

Triage pending correction candidates (captured by the recall Stop hook) into memory — promote the worthy, drop the noise, surface the contested. Closes the correction→memory loop.

First-run personalization for a freshly cloned Lab — interview the user one question at a time and write their IDENTITY.md, so the agent knows who it's working with. Use on first run, when IDENTITY.md still has placeholders, or when the user says "/setup", "set up my lab", or "personalize this".

Cleanly wrap a work session — ground state against git+filesystem, update and de-bloat the tracking files, ensure the next-session landing pad is grounded (not invented), scan staged changes for leaked secrets, then commit (and push if there's a remote). Invoke at session end, or whenever the user says "wrap".

Drift audit for tracking files, specs, plans, or citations against live state. Produces one Reviews/ doc with mechanically-verified findings, two-pass (inventory then apply).

Re-explain your own previous answer in plain language, for when it didn't land. Use when the user types /bro, or says the last answer was too dense, too jargon-heavy, or asks for it simpler. Re-expresses what was already said; never answers something new.

Periodic sweep of the workspace's curated narrative (session-note titles + git subjects + roadmap/DECISIONS) to surface repeated, formalizable work and propose turning it into skills or scripts. Writes one Reviews/ doc with a verdict per candidate. NOT correction triage (that's /review-corrections) and NOT drift detection (that's /audit) — this finds un-formalized recurring WORK.

Run a structured planning session — interview the sources of truth, draft, adversarial review (up to 3 reviewers, scaled to the change), approve, update docs. Required for any change to the shipped engine, a skill/template, a hook, recall, security, or a public release, and for features spanning several files or systems; a small change touching none of those doesn't need it.

Triage pending correction candidates (captured by the recall Stop hook) into memory — promote the worthy, drop the noise, surface the contested. Closes the correction→memory loop.

Cleanly wrap a work session — ground state against git+filesystem, update and de-bloat the tracking files, ensure the next-session landing pad is grounded (not invented), scan staged changes for leaked secrets, then commit (and push if there's a remote). Invoke at session end, or whenever the user says "wrap".

Drift audit for tracking files, specs, plans, or citations against live state. Produces one Reviews/ doc with mechanically-verified findings, two-pass (inventory then apply).

Re-explain your own previous answer in plain language, for when it didn't land. Use when the user types /bro, or says the last answer was too dense, too jargon-heavy, or asks for it simpler. Re-expresses what was already said; never answers something new.

Periodic sweep of the workspace's curated narrative (session-note titles + git subjects + roadmap/DECISIONS) to surface repeated, formalizable work and propose turning it into skills or scripts. Writes one Reviews/ doc with a verdict per candidate. NOT correction triage (that's /review-corrections) and NOT drift detection (that's /audit) — this finds un-formalized recurring WORK.

Run a structured planning session — interview the sources of truth, draft, adversarial review (up to 3 reviewers, scaled to the change), approve, update docs. Required for any change to the shipped engine, a skill/template, a hook, recall, security, or a public release, and for features spanning several files or systems; a small change touching none of those doesn't need it.

Triage pending correction candidates (captured by the recall Stop hook) into memory — promote the worthy, drop the noise, surface the contested. Closes the correction→memory loop.

Cleanly wrap a work session — ground state against git+filesystem, update and de-bloat the tracking files, ensure the next-session landing pad is grounded (not invented), scan staged changes for leaked secrets, then commit (and push if there's a remote). Invoke at session end, or whenever the user says "wrap".