Skills
Every slash command Strata ships. Some are auto-invokable: when you describe a situation that matches the skill’s description, Claude triggers it without you typing the command. Others are user-only. They take destructive or audit-sensitive actions.
Writes (auto-invokable)
Section titled “Writes (auto-invokable)”| Command | When it fires | Output |
|---|---|---|
/strata:save <topic> | ”Save this”, “capture what we just discussed”, Stop hook nudge | pr-context/<branch>/...md |
/strata:decide <title> | ”We’ve decided X”, “Let’s commit to Y” | decisions/YYYY-MM-DD-<slug>.md |
/strata:domain <concept> | A domain term is defined or refined (“an Order always belongs to one Customer”) | domain/<slug>.md |
/strata:correct <note> | ”Fix the part about X”, “Stop using | Updates note + audit log |
/strata:lint | Before commit, automatic via pre-push hook | Scans for secrets / PII |
/strata:review | Asked to audit vault health | Surfaces stale-proposed ADRs, decayed durable notes (rarely recalled), orphans, drift |
/strata:doctor | ”Is Strata working?”, “did the install work?”, after /strata:init | Health check: runtime deps, vault, index, semantic search, MCP server, git |
/strata:audit-config | Asked about plugin config | Reports active settings |
Reads (auto-invokable)
Section titled “Reads (auto-invokable)”| Command | When it fires |
|---|---|
/strata:find <terms> | ”Search for…”, “do we have notes about…” |
/strata:resume | New session, surfaces last branch + latest notes |
Most reads happen via MCP rather than skill invocation. Claude calls recall (the memory-recall path) automatically when conversation overlaps known topics. See MCP tools.
User-only (high stakes)
Section titled “User-only (high stakes)”| Command | What it does |
|---|---|
/strata:forget <note> | Move note to .trash/ with audit log. Reason required. |
/strata:archive | Move merged-PR context to archive folder |
/strata:export-to-repo | Copy vault notes into a repo for committing |
/strata:promote-to-pr | Surface vault context in PR description |
These never auto-invoke. The audit trail must reflect the user’s explicit intent.
Setup / lifecycle
Section titled “Setup / lifecycle”| Command | Use |
|---|---|
/strata:init | One-time vault setup for a repo |
/strata:bootstrap | One-time migration of existing planning docs (bootstrap guide) |
/strata:graphify | Build / refresh the code graph |
Auto-invocation rules
Section titled “Auto-invocation rules”A skill auto-fires when:
- Its
description:frontmatter matches the user’s intent pattern - The action is safe and reversible (writes to new files; not destructive to existing data)
- The user hasn’t explicitly typed a slash command for a different skill
If you want Claude to STOP auto-invoking a skill for a turn, prefix your message with “don’t save this” or similar. Claude respects opt-out phrasing.
Composability
Section titled “Composability”Skills call each other through skill instructions, not Python imports. /strata:bootstrap dispatches strata:bootstrap-worker subagents that invoke /strata:decide / /strata:domain / save scripts internally. Each layer stays composable.
Next: Bootstrap — the one-time migration that pulls existing planning docs into the vault.