| name | doc-init |
| description | Start (or resume) the conversational documentation build. Invokes the doc-orchestrator agent which audits the current docs/ state, identifies the current phase + next gap, and begins / resumes the discovery interview. Idem- potent — safe to call repeatedly. Use when a project has just been scaffolded with `pentaglyph init` or when picking up after a long pause.
|
| disable-model-invocation | false |
/doc-init — start (or resume) conversational doc build
Use this skill when you want to build (or continue building) the project's
docs/ tree through conversation with the doc-orchestrator.
What it does
- Hands off to the
doc-orchestrator agent
- The orchestrator runs the
completeness-auditor to see current state
- The orchestrator decides the current phase (1 Discovery / 2 Architecture
/ 3 Detail / 4 Operations / complete) and shows you the coverage report
- The orchestrator asks the next focused question (one at a time)
- As answers accumulate, the orchestrator dispatches specialist agents
(
discovery-agent, architect-agent, adr-writer, spec-writer) to
write each artefact
- After every dispatch, the auditor re-runs and the orchestrator re-decides
The conversation continues until the auditor reports 100% coverage.
When to use
- Fresh project: just ran
pentaglyph init, the docs/ scaffold has only
README.md files. → /doc-init to start Phase 1 (Discovery)
- Resuming: a prior conversation got partway through. →
/doc-init again;
the orchestrator picks up exactly where you left off (state is computed
from disk, not stored anywhere)
- Recovering from a hand-edit: you edited some docs by hand and want to
know what's still left. →
/doc-init will re-audit and tell you
When NOT to use
- You only want a status check, no conversation → use
/doc-status instead
- You know the specific gap you want to fill → use
/doc-fill <area> instead
- You're documenting a single bug fix or one-shot change → just talk to
Claude directly; no orchestration needed
Output
The orchestrator's first turn always includes a coverage table like:
PHASE: 2
COVERAGE: 35%
SUBSTANTIVE-FILES: 8 / 22
ARTEFACTS:
PHASE-1:
✓ docs/arc42/01-introduction-and-goals/overview.md: 7 goals, 4 stakeholders, 5 quality goals
✓ docs/arc42/03-context-and-scope/system-context.md: 2 actors, 3 external systems
✓ docs/arc42/03-context-and-scope/prds/auth-prd.md: 4 FRs, 3 NFRs
✗ no use cases yet
PHASE-2:
⚠ docs/arc42/04-solution-strategy/strategy.md: only 2 decisions, need ≥ 5
✗ docs/arc42/05-building-blocks/overview.md: missing
✗ docs/diagrams/c4/workspace.dsl: still in starter state
⚠ docs/arc42/09-decisions/: only 1 ADR, need ≥ 3
NEXT-GAP: write a use case for the primary auth flow
NEXT-GAP-PATH: docs/arc42/03-context-and-scope/use-cases/login.md
NEXT-GAP-TEMPLATE: 4 (Use Case)
NEXT-GAP-AGENT: discovery-agent
NEXT-GAP-MISSING: actor, main success scenario steps, ≥ 1 Given/When/Then AC
Followed by a single focused question targeting that gap.
Implementation note
This skill is a thin pointer. The actual logic lives in
.claude/agents/doc-orchestrator.md. Edit the orchestrator's prompt to
change behaviour; this skill file should not need updates.
See also