pg-implementlisted
Install: claude install-skill matejformanek/postgres-claude
# pg-implement — Phase 3 of the PG planner suite
The third stage. Brainstorm narrowed the design space; Plan made it
implementable; **Implement walks the plan phase-by-phase** while
enforcing the discipline rules.
The pairing:
- Phase 1 — `pg-feature-brainstorm` (sketch)
- Phase 2 — `pg-feature-plan` (heavy plan)
- **Phase 3 — `pg-implement`** (this skill, executes the plan)
## When to use vs the generic `/implement`
| Project | Use |
|---|---|
| Anything PG-related with a `planning/<slug>/plan.md` | **`/pg-implement`** |
| Generic project plan, not PG | Generic `/implement` |
| Ad-hoc PG coding, no plan | Neither — write a plan first via `/pg-plan` |
The PG variant exists because PG implementation has unusual
constraints the generic `/implement` doesn't enforce:
- Per-phase test runs against the dev cluster
- File:line citation discipline (knowledge corpus must stay accurate)
- Plan-linked commit messages (commits reference the plan + phase)
- Upstream-vs-meta commit-message-style split
- Catalog/catversion/WAL-format pre-flight before any touching edit
- `/pg-restart` cadence after backend code changes
## Inputs
- **Slug** (required): the planning directory under `planning/<slug>/`.
Examples in this repo: `sp2-pgstr-maxalloc`, `cb1-pgcrypto-bomb`,
`sp7-tablefunc-quoting`. The `dev/` branch mirrors the slug
(e.g. `feature_sp2_pgstr_maxalloc`, `feature_server_side_vars`).
- Must contain `plan.md` produced by `pg-feature-plan`.
- May contain `brainstorm.md` (Phas