Runbook

Runbook Generation Skill

Trigger

• Keywords: runbook, release runbook, deployment handbook, release handbook, operational guide, pre-release checklist, rollback plan

When NOT to Use

Usage

/runbook                              # Auto-detect feature, create or update
/runbook <feature-keyword>            # Specify feature
/runbook --update                     # Force update mode
/runbook --check                      # Read-only staleness validation
/runbook --request <path|title>       # Specify target request (multi-request features)

Workflow

sequenceDiagram
    participant U as User
    participant S as /runbook
    participant FR as Feature Resolver
    participant CB as Codebase
    participant RB as runbook-release.md

    U->>S: /runbook [feature] [--update|--check] [--request path]
    S->>FR: node scripts/resolve-feature.js
    FR-->>S: {key, doc_inventory, source sets}
    S->>S: Mode dispatch + Request selection

    alt Create Mode
        S->>CB: Read current_authority + requests/*.md
        S->>CB: Scoped discovery (5-priority cascade)
        S->>RB: Write runbook-release.md from template
    else Update Mode
        S->>RB: Read existing runbook + provenance
        S->>CB: Compare current state vs provenance SHAs
        S->>RB: Edit changed sections only
    else Check Mode
        S->>RB: Read existing runbook + provenance
        S->>CB: Validate per-section SHAs
        S-->>U: Report: Fresh/Stale/Missing/Unknown
    end

Phase 0: Context Resolution

Resolve feature using the 5-level cascade:

The wrapper, not the CLI directly: resolve-feature.js owns the failure payload, so the full shape with scan_error: true arrives however the CLI fails — a nonzero exit, a signal, a partial write, a payload that is not the agreed shape. (It cannot survive node itself being unavailable: nothing running under node can. What it removes is the CLI's failure domain, not the interpreter's.) Calling the CLI with || echo '{}' produces a payload the gate below cannot recognise as a failure.

Decide the branch yourself, then run one command. This skill grants Bash(node:*), which matches a direct node … invocation and nothing else — a shell if/[ … ]/$(…) compound is not a node command and cannot run here. Parse $ARGUMENTS first (Step 1 below), then issue exactly one of:

node scripts/resolve-feature.js --feature <the feature key from $ARGUMENTS>

node scripts/resolve-feature.js

Use the first when $ARGUMENTS carried a positional feature key, the second otherwise. Pass the key as a separate argv token — never interpolate it into a larger shell expression.

scan_error gate. Gate on scan_error !== false, not on scan_error === true. When it is not exactly false the four source sets are unknown, not empty — the corpus could not be enumerated (unreadable directory, broken taxonomy, no repository), or the resolver never ran and a shell fallback supplied a payload with no such field at all. {} is the shape that made the stricter test useless: it has no scan_error, so === true is false and the gate passes a payload that contains nothing. Do not proceed as though the feature has no authority documents — report and take the ⚠️ Need Human exit. A key may still be present, so a non-null key is not evidence the sets are complete.

Note: Mode dispatch keys off the specific file runbook-release.md, not any runbook-typed doc in doc_inventory. A feature may have runbook-deploy.md (a different topic) without triggering update mode for the release runbook.

Request Selection

Phase 1: Content Discovery (Create/Update modes)

Use scoped discovery cascade — narrow to wide, with confidence degradation:

A P1 path is classified before it is used. Related Files is High confidence because the request author named those paths deliberately — not because a path in that table is exempt from the role split. Resolve each one first: a path landing in design_records (a tech spec, an architecture doc) is treated as P3 — Medium, marked unverified — even though it arrived via P1. Otherwise the row the split removed comes straight back through the front door, since a request's Related Files table routinely names 2-tech-spec.md.

Priorities 2 and 3 used to be one row reading "canonical docs (tech-spec, architecture) — High", which is the confusion this feature exists to remove: a tech spec is a design record, and a runbook built from one describes a procedure that may never have been built.

See references/discovery-heuristics.md for per-section mapping.