@skitterbyte/skitterspec 18.0.0 → 20.0.0

package/assets/skills/spec-reviewed/SKILL.md

@@ -0,0 +1,258 @@
+---
+name: spec-reviewed
+description: Pick up the review you approved on the page — run it bare to pick up the single waiting pass, paste the six-digit code off the page ("/spec-reviewed 608223") to name one exactly, and be asked which only when two are waiting. Use when the user says "/spec-reviewed", "I approved it", "I've reviewed it", "pick up my review", "I pressed approve", or otherwise says they have finished reviewing a rendered diff.
+disable-model-invocation: true
+---
+
+# /spec-reviewed — pick up the review you just approved
+
+> Stay silent while this runs — speak only to ask something you cannot answer
+> yourself, or to report a failure at the moment it happens. Read
+> `.claude/rules/spec-reports.md` before reporting; it defines the block this
+> skill ends with.
+
+You read the diff somewhere else, pressed a verdict, and the engine is holding
+the pass. This is how it gets picked up.
+
+## What this is for, now that the picker exists
+
+`/spec-next` and `/spec-diff` both end in a picker carrying these same endings,
+so the usual way to pick a review up is to choose one there. This command is
+what covers the case that picker cannot.
+
+**A picker is consumed when the turn ends.** Nobody scrolls back and answers one
+an hour later. So a review read over lunch, a session cleared, a fresh terminal
+tomorrow morning — in every one of those the control is gone and the pass is
+still waiting, and this is the way in. That is a narrower job than it once had,
+and not an absent one.
+
+**It is also the only claim path the harness itself enforces** — see below.
+Deleting this skill as redundant with the picker would leave no path that has
+the mechanism at all, only paths that have the property.
+
+## Why this is user-only, and why that is the whole point
+
+`disable-model-invocation: true` is not ergonomics here — it is the enforcement
+of `/spec-diff` step 0's central rule:
+**never claim a pass you were not asked to claim**.
+
+That rule exists because a review pass can be POSTed by anything that reaches
+the page. So a stranger's pass sits in the holding area — unless an agent goes
+and fetches it, which is exactly what happened once: an agent found a waiting
+approval, read its code off disk, claimed it, and reported the round-trip
+working.
+
+Prose alone did not prevent that. The harness does —
+**the model cannot invoke this skill** — so a pass picked up *this way* is only
+ever picked up because a person typed the command. Typing it *is* the human
+signal. A later edit that makes this skill model-invocable does not make it more
+convenient — it removes the only thing standing between a stray approval and
+someone's repo.
+
+**There is exactly one other way in, and it is bounded rather than trusting.**
+A phase that ends waits on its page, and a pass arriving *during that wait* is
+claimed by the engine (`--claim-since`, `/spec-diff` §4b): scoped to the window,
+acting on nothing when no pass arrived, refusing when two did. A pass already
+sitting there when the wait began is never swept up by it — which is the
+stranger's pass this whole rule was written about. Everything outside that
+window still comes through here.
+
+## 1. Resolve the spec
+
+Bare, exactly as every other bare command in this workflow: the worktree you are
+standing in, else the sole provisioned spec. Ask the engine rather than guessing:
+
+```
+skitterspec spec-env resolve
+```
+
+Several provisioned and none resolved is a refusal — relay its list and stop,
+never pick from it. See `.claude/rules/spec-planning.md`; do not restate the
+rule here.
+
+**Three argument shapes, and they cannot collide.** A **six-digit code**
+matches `^\d{6}$`; a **tracker id** carries a letter and a hyphen; a
+**spec name** is neither. So the parse needs no flag, and nothing has to be
+guessed at from context.
+
+A **six-digit code** is a pass the operator read off their own page. It says
+**which pass**, not which spec — so resolve the spec exactly as a bare
+invocation does, above, and claim the code *there*. Then go to step 4 — a named
+pass has nothing to disambiguate.
+
+A **name** targets that spec instead. A **tracker id** does too, but only
+through a **provider seam**: the base knows nothing about tracker ids, so
+resolve one by asking whichever provider is installed for its listing —
+
+```
+skitterspec spec-sync linked --json
+```
+
+— which answers `[{ spec, bucket, identifier }]`, and match on `identifier`.
+**With no provider installed an id resolves to nothing**, and that is the right
+answer rather than a guess: a `SKS-227`-shaped string is not evidence that a
+tracker exists. Say the id matched no spec and stop.
+
+## 1a. Targeting another spec? Get into its worktree first
+
+**Only when the resolved spec's worktree is not where this session stands.**
+Compare the `worktree:` line from `skitterspec spec-env resolve <spec>` against
+cwd, resolving both paths first so a symlinked or trailing-slash spelling of one
+tree does not read as two. **Same tree — say nothing and carry on.** That is the
+ordinary case, and a line about it is narration.
+
+**Different trees, and it matters for what comes after, not for the claim.**
+Claiming is harmless from anywhere: the sidecar lives in the primary checkout.
+But an honoured `commit` verdict **commits**, and `changes` **edits files**, and
+both must land in *that spec's* worktree — done from here they would land in
+this one, on the wrong branch, looking entirely normal at the time.
+
+So **ask, then move**:
+
+*"`feat-orders` lives in `../repo-wt/orders`, and picking its review up means
+committing there — move this session over to carry on?"*
+
+- **On a yes**, move with a plain `cd "<worktreePath>"`. That is the whole
+  mechanism, exactly as `/spec-start` does it — not a tool call, because an
+  approval prompt is unusable on a phone.
+  Then **confirm the move landed** rather than assuming it
+  (`.claude/rules/negative-checks.md` rule 1): run `skitterspec spec-env resolve`
+  with no argument and check its `spec:` line names the target. If it does not,
+  stop — do not claim a pass you are about to act on from a tree you could not
+  confirm.
+- **On a no, stop without claiming.** A pass claimed here and acted on there is
+  exactly the split this guard exists to prevent, and claiming first would spend
+  the code for nothing.
+
+**A target with no worktree is a refusal.** There is nowhere for a commit to
+land. Name it, suggest `/spec-start <name>`, and stop.
+
+**Why not simply refuse unless you are on the base branch?** Because that bans a
+legitimate case — standing in one spec, picking up another's review — while
+*still* leaving the work to be done in a tree you are not in. The relocation is
+needed either way, so the relocation is the guard.
+
+## 2. Read what is waiting
+
+```
+skitterspec spec-env review <spec>
+```
+
+Its `pending:` block lists each waiting pass — code, verdict, age:
+
+```
+pending: 1 waiting
+  792969 · commit · 1 min ago
+```
+
+**Never open `.spec-env/reviews/<spec>.pending.json`.** The render carries
+everything a decision needs, so there is nothing to go looking for, and going
+looking is the bypass step 0 forbids. The file is right there and readable; that
+is precisely why the rule is written down rather than assumed.
+
+**Check the store before concluding nothing is waiting.** Only when this spec
+has a published page (`--json` reports a `url`): a reader off the LAN sent their
+pass to the artifact's own `passes` collection rather than to the engine, so an
+empty holding area is not an empty answer yet. Read it with the Artifact tool's
+`read_db`, merge the `blob` through `--notes`, and **delete the document** —
+`/spec-diff` §6 owns the whole sequence, including that a claim consumes.
+
+**Nothing waiting in either is an ordinary answer.** Say so and stop. Do not hunt
+through other specs, and do not treat an empty holding area as a problem — a
+`file://` page copies to the clipboard instead of sending, so mention that the
+pass may be on their clipboard waiting to be pasted, and finish.
+
+## 3. One pass waiting? Act on it
+
+**Claim it and go to step 4.** Do not read the code out, and do not ask whether
+it is theirs. The operator typed this command, which is the whole signal — and
+this skill is the one place in the workflow where that signal cannot be
+manufactured, because the model cannot invoke it.
+
+**The code was never an authorisation.** A later edit must not restore it as
+one. It earned its keep when the *page* pushed: the agent went looking, found
+a pass, and had to prove which one it had. What stops a stranger's approval
+reaching your repo is not those six digits — it is that this command cannot be
+typed by anything but a person, and that the one automatic path (`--claim-since`)
+is bounded by a window a stranger's pass falls outside of. The confirmation step
+added nothing to either, and cost a round-trip on every review.
+
+So the code has exactly one job left: **telling two passes apart**. That is
+disambiguation, not a gate.
+
+**Two or more waiting is a refusal to guess.** Name them all — code, verdict,
+age — and ask which. Never take the newest, the oldest, or the only `commit`:
+this is the one case where a stranger's pass really is sitting beside theirs,
+and the six digits are the only thing that separates them.
+
+*"Two are waiting — 792969 (commit, 1 min ago) and 324199 (commit-continue,
+just now). Which is yours?"*
+
+**A pasted code skips even that.** `/spec-reviewed 324199` names the pass
+outright, so there is nothing to disambiguate and nothing to ask — claim it and
+act.
+
+**On "that isn't mine"**, leave it and offer to drop it:
+
+```
+skitterspec spec-env review <spec> --drop <code>
+```
+
+A pass that stays is reported on every render until the operator stops reading
+the line — which is how the real one gets missed.
+
+## 4. Claim it and act
+
+```
+skitterspec spec-env review <spec> --claim <code>
+```
+
+Then **route on the verdict exactly as `/spec-diff` §2 does** — its steps 2, 2a
+and 4 own that routing, including the commit hand-off through
+`review.commitWith` and what `commit-continue` does after it. Follow it; do not
+restate it here, because two copies of a routing rule is how the two come to
+disagree.
+
+That includes the one rule worth knowing before you invoke anything:
+`commit-continue` runs `/spec-next` and **stops there**. It never completes,
+lands or tears anything down.
+
+**A code that matches nothing refuses, and names nothing.** The engine answers
+that way and you relay it unchanged: do not list what *is* waiting, and never
+fall back to "the only one". Both would hand a guesser the answer, and the
+second is exactly the fallback that would let an unread pass through. A mistyped
+digit is the ordinary cause — say the code matched nothing and let the operator
+look again.
+
+That holds whichever door the code came through, the paste included. A wrong
+code is a wrong code.
+
+A claim is a **delivery mechanism, not a second kind of review**: a pass that
+arrived this way means exactly what the same pass pasted into the chat would
+mean — or picked from `/spec-next`'s picker, which is the same conclusion
+reached by a third route.
+
+## 5. Report
+
+End with the block defined in `.claude/rules/spec-reports.md`. That file carries
+the shape; this section carries only what is specific here.
+
+**Verdicts**
+
+- `✅` — a pass was claimed and its verdict acted on.
+- `⚠️` — claimed and acted on, with something worth knowing.
+- `❌` — it acted and stopped part-way; the commit failed, or the work did.
+  Quote it.
+- `⏸` — nothing was waiting, the operator said the waiting pass was not theirs,
+  they declined the move to another spec's worktree, or the target has no
+  worktree at all. Nothing changed, and none of those is a failure.
+
+**Fields:** `Tracker` · `Branch` · `Built` · `Tests` · `Worktree` ·
+`Follow-ups` · `Next`
+
+`Worktree` appears only when this run **moved the session** — the path it moved
+to, because the operator's next command depends on knowing where they now are.
+A run that stayed put omits it.
+
+`Built` is what the verdict produced — the commit, or the commented files
+worked. A run that claimed nothing built nothing and omits it.