plumbbob 0.2.3 → 0.3.1

package/skills/pb-start/SKILL.md

@@ -1,19 +0,0 @@
----
-name: pb-start
-description: Human-triggered driver for `plumbbob start` — scaffold a new session (.plumbbob/, STATE=DESIGN, baseline commit) without leaving the chat.
-disable-model-invocation: true
-model: haiku
-allowed-tools: Bash(__PLUMBBOB_BIN__ status:*), Bash(__PLUMBBOB_BIN__ start:*)
----
-
-# Plumbbob — start a session (driver)
-
-Current session state (injected when this skill runs): !`__PLUMBBOB_BIN__ status`
-
-This is a **driver skill** — a chat-side trigger for the mechanical `plumbbob start` verb, so the whole workflow runs from the agent window instead of a terminal. It is `disable-model-invocation: true`: only the human fires it. It carries **no Edit and no Write tool** — its only action is to shell the verb and report the verb's output verbatim, including any refusal. The CLI is the source of truth: never retry a refused transition, and never edit a file to work around one.
-
-## What it does
-
-1. Read the session title from the way you were invoked (e.g. `/pb-start "fix the widget"` → title `fix the widget`). If no title is present, ask for one and run nothing.
-2. Run `__PLUMBBOB_BIN__ start "<title>"` via Bash.
-3. Report the verb's output verbatim. If it refuses (e.g. a session already exists), relay that and stop.

package/skills/pb-wrap/SKILL.md

@@ -1,18 +0,0 @@
----
-name: pb-wrap
-description: Human-triggered driver for `plumbbob wrap` — set STATE=FINISH so the report and docs skills can run.
-disable-model-invocation: true
-model: haiku
-allowed-tools: Bash(__PLUMBBOB_BIN__ status:*), Bash(__PLUMBBOB_BIN__ wrap:*)
----
-
-# Plumbbob — wrap into FINISH (driver)
-
-Current session state (injected when this skill runs): !`__PLUMBBOB_BIN__ status`
-
-This is a **driver skill** — a chat-side trigger for the mechanical `plumbbob wrap` verb, so the whole workflow runs from the agent window instead of a terminal. It is `disable-model-invocation: true`: only the human fires it. It carries **no Edit and no Write tool** — its only action is to shell the verb and report the verb's output verbatim, including any refusal. The CLI is the source of truth: never retry a refused transition, and never edit a file to work around one.
-
-## What it does
-
-1. Run `__PLUMBBOB_BIN__ wrap` via Bash.
-2. Report the verb's output verbatim. FINISH is what unlocks `/plumbbob-report` and `/plumbbob-docs` — next, run `/plumbbob-report`.

package/skills/plumbbob-docs/SKILL.md

@@ -1,25 +0,0 @@
----
-name: plumbbob-docs
-description: FINISH-phase, optional docs update — conservatively project canonical intent into real docs/, only when warranted.
-disable-model-invocation: true
-model: opus
-allowed-tools: Read, Edit, Write, Bash(__PLUMBBOB_BIN__ status:*)
----
-
-# Plumbbob — update the docs
-
-Current session state (injected when this skill runs): !`__PLUMBBOB_BIN__ status`
-
-## Wrong-state refusal
-
-`docs/` is writable **only in FINISH** (D19) — the muzzle blocks it everywhere else. Read the state injected above and **refuse if it is not `FINISH`**, naming `plumbbob wrap` as the verb that enters FINISH (the report skill runs first; this docs pass is the optional second Finish step). If the state is `NO ACTIVE SESSION` the hooks are dormant and there is nothing for this skill to gate — just edit docs the ordinary way.
-
-## What this skill does
-
-**Conservative by default.** Most sessions write no docs at all — **a bug fix usually shouldn't spawn a doc**. Update real documentation under `docs/` only when the work genuinely changed how the system should be described, and only from the **canonical** parts of `intent.md` (the Frame, the Decisions) — never from the chat transcript.
-
-## The one hard contract
-
-- Default to doing nothing. Open a doc edit only when you can name what changed for a reader of `docs/`.
-- Project only from `intent.md`; the report is for the session, the docs are for the system.
-- Stay inside `docs/` — and only in FINISH. An edit anywhere else, or in any other state, is the muzzle's job to refuse, not yours to attempt.

package/skills/plumbbob-report/SKILL.md

@@ -1,35 +0,0 @@
----
-name: plumbbob-report
-description: FINISH-phase report — synthesize intent + build-log into exactly .plumbbob/report.md with the five required sections.
-disable-model-invocation: true
-model: opus
-allowed-tools: Read, Write, Bash(__PLUMBBOB_BIN__ status:*)
----
-
-# Plumbbob — write the report
-
-Current session state (injected when this skill runs): !`__PLUMBBOB_BIN__ status`
-
-## Wrong-state refusal
-
-This skill writes the report **in FINISH only**. Read the state injected above:
-
-- `STATE: FINISH` — proceed.
-- Any other active state (`DESIGN`, `BUILD`, `REVIEW`, `SPIKE`) — **stop and tell the human to run `plumbbob wrap` first** (D28). Entering FINISH is what unlocks writing `report.md`, and this skill is the thing that tells you to wrap.
-- `NO ACTIVE SESSION` — there is nothing to report; start a session with `plumbbob start "<title>"`.
-
-## What this skill does
-
-The first of the three Finish steps. Read `intent.md` and `build-log.md` and synthesize the conclusion — the "yeah, I did that" artifact — into **exactly one file: `.plumbbob/report.md`**. Write nothing else, and touch no other path.
-
-## Required sections
-
-`.plumbbob/report.md` must contain these five sections, in order:
-
-1. **What shipped** — what this session actually built, measured against the Frame's "done looks like".
-2. **The decisions and why** — the decisions taken (the `D#` ledger) and the reasoning behind each.
-3. **Parked items and how each was triaged** — every Park-list item and its triage outcome (blocker / tangent / pivot signal).
-4. **Final status** — done, partial, or abandoned, with the checkpoint SHAs.
-5. **Deferred tangents (future Plumbbobs)** — the tangents worth their own future session.
-
-Once `report.md` exists, the human runs `plumbbob finish` (the closing gate refuses without it) to archive and clear the session.

package/skills/plumbbob-triage/SKILL.md

@@ -1,38 +0,0 @@
----
-name: plumbbob-triage
-description: DESIGN-phase, step-boundary triage — propose one class per parked item, write only after the human confirms each.
-disable-model-invocation: true
-model: opus
-allowed-tools: Read, Edit, Bash(__PLUMBBOB_BIN__ status:*)
----
-
-# Plumbbob — triage the park list
-
-Current session state (injected when this skill runs): !`__PLUMBBOB_BIN__ status`
-
-## Wrong-state refusal
-
-This skill runs in **DESIGN only**, at a **step boundary** (after the step's check has gone green). Read the state injected above and **stop before writing anything** if it is not `DESIGN`:
-
-- `STATE: BUILD` or `STATE: REVIEW` — finish the step first with `plumbbob done`, which returns you to DESIGN; then triage.
-- `STATE: SPIKE` — close the experiment with `plumbbob spike done` first.
-- `STATE: FINISH` — too late to triage; the report skill folds the park list in instead.
-- `NO ACTIVE SESSION` — start one with `plumbbob start "<title>"`.
-
-When the state is wrong, refuse in one line naming the verb above.
-
-## What this skill does
-
-Walk the **Park list** in `build-log.md` item by item and, for each, **propose exactly one class**:
-
-- **blocker** — the plan was wrong or incomplete; can't proceed. Folds into `intent.md` and is handled now.
-- **tangent** — a different path, not clearly better. **The default** — defer or kill.
-- **pivot signal** — real evidence the whole approach is wrong. Stop and replan deliberately.
-
-## The one hard contract
-
-You **propose**; the **human calls it** (D13). For each item, state your proposed class and one line of reasoning, then **wait for the human to confirm or override**. Write **only after** per-item confirmation:
-
-- Record each confirmed class in `build-log.md`'s `## Triage` section.
-- A confirmed **blocker** also folds its decision into `intent.md`.
-- Never reclassify or resolve an item the human hasn't confirmed, and default every uncertain item to **tangent**, never to blocker.