Init Skill for Claude

/prd-ce:init — Greenfield Scaffold Seeder

Plant the consumer-owned scaffold for a fresh PRD-CE project. The methodology engine

(lifecycle skills, the agent squad, governance hooks, readiness.py) is delivered live

by the installed prd-ce plugin — it never gets copied into your repo. This skill seeds only

what is yours to own and edit: PRD.md, the SoT/ graph, epics/ templates, your

.claude/domain-profile.yaml, and per-agent MEMORY.md starters.

> One manifest, no drift. The seed list comes from install-manifest.yaml's

> template_seed section — the same list install.sh (fork path) and ghm-template-sync

> (update path) read. This skill drives the deterministic prd-ce-init.sh so behavior is

> identical no matter who invokes it.

> Scope (v1): greenfield only. This seeds an *empty* structure into a fresh repo.

> Mid-build and live-codebase on-ramps (entry-mode branching, graph extraction) are

> backlogged — see temp/plugin-conversion-plan.md. If the target already has PRD.md or

> SoT/ content, the seeder keeps it (non-destructive) rather than adapting to it.

Consumes

• ${CLAUDE_PLUGIN_ROOT}/templates/ — the bundled seed sources (mirrors template_seed

paths) the packager ships with the plugin.

• install-manifest.yaml template_seed + never_touch — authoritative seed/skip lists.
• scripts/prd-ce-init.sh — the deterministic seeder this skill drives.

Produces

• A seeded scaffold in the target repo (no new SoT IDs — this skill *places* templates, it

does not author specs).

• A freshly reset PRD.md (frontmatter at v0.1, today's date, no stale template_version).
• A verification report (hooks emit valid JSON; readiness.py runs).

Workflow

Phase 1 — Preflight

• Confirm git, python3, bash are present. Warn (don't block) if the target isn't a git repo.
• Confirm greenfield: if PRD.md or non-empty SoT/ already exist, say so and stop —

the seeder will keep them untouched, so there is nothing for init to do. (Point the user at

the lifecycle skills to keep building, not at re-seeding.)

Phase 2 — Wizard questions

Ask only what changes the outcome (honor the execution mode's budget):

• Target directory (default: current repo).
• Domain profile: product (default) · library · infrastructure · research.

Quick mode skips this and takes the default.

Phase 3 — Seed (drive prd-ce-init.sh)

Run the deterministic seeder so behavior matches every other path:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/prd-ce-init.sh" --target <DIR> --dry-run   # preview
bash "${CLAUDE_PLUGIN_ROOT}/scripts/prd-ce-init.sh" --target <DIR>             # seed

• Show the --dry-run plan first [standard+], then execute.
• The seeder skips any file that already exists (never_touch honored) and resets a freshly

seeded PRD.md frontmatter to v0.1.

• After seeding, if the chosen domain profile differs from the default, update the

profile: key in the seeded .claude/domain-profile.yaml [standard+].

Phase 4 — Verify (trust-but-verify)

• Run each plugin hook against the target; assert valid JSON on stdout.
• Run python "${CLAUDE_PLUGIN_ROOT}/scripts/readiness.py" run — a BLOCK on an empty

scaffold is the gate working (no content yet), not a failure. Report the score.

• Print next steps: customize README.md + PRD.md, then "Let's frame the problem" (v0.1).

Anti-patterns

| Pattern | Fix |

|---------|-----|

| Hardcoding the seed file list in the skill | Drive prd-ce-init.sh; it reads template_seed |

| Copying the framework (skills/hooks) into the consumer repo | The plugin provides those live — seed only consumer-owned files |

| Overwriting an existing PRD.md/SoT/ | Greenfield-only; the seeder skips what exists |

| Leaving PRD.md at the example's version header | Frontmatter reset to v0.1 happens automatically on a fresh seed |

| Treating a readiness BLOCK on a fresh scaffold as a bug | It's the gate working — report the score |