Starter kits

spec-dev-review — the adversarially-reviewed spec packet

For one ticket or story, writes a scoped spec packet — findings, progress, task plan, rubric — and rounds it through a genuinely independent…

For one ticket or story, writes a scoped spec packet — findings, progress, task plan, rubric — and rounds it through a genuinely independent adversarial review that regression-checks prior findings, not just new ones. Stops on a clean pass or logs the remaining blocker for a human.

Category: D · Issue & intake · Heartbeat: conditional (run-until-done, on-demand — one per ticket) · Level: L1 (report-only, drafts-not-final) · Cost: Medium

Source: Loop #21, "The spec dev-review loop," from Forward Future's Loop Library, original prompt by Ximanta (@kharamdau).

Quickstart

1. Read LOOP.md — the six parts, the prompt, the limits, the three stops,
   and the source's verbatim prompt.
2. spec-dev-review-state.md is your spine — already committed with an
   empty review log.
3. Claude Code: run it in-session on a real ticket, pointed at
   .claude/skills/loop-task/SKILL.md. OpenCode: rename
   opencode.json.example → opencode.json, add the AGENTS.md line, run
   skills/loop-task.md.
4. Rehearse once, by hand, on a real (small) ticket before trusting it on a
   big one: "Run the spec-dev-review skill now on [ticket]."
5. Grade the rehearsal with the loop-verifier agent — it fails any "ready"
   verdict that missed a regressed prior finding. PASS → trust it on real
   tickets. FAIL → fix the skill, never the packet.

What's in this kit

File	Role
`LOOP.md`	the definition — six parts, prompt, limits, ownership, three stops
`spec-dev-review-state.md`	the spine — round count, the review log (finding + fix per round)
`loop-budget.md`	caps: 5 rounds/ticket, 300k tokens — the round number is this library's addition
`loop-constraints.md`	the constitution — reviewer must be genuinely independent, prior findings must stay resolved
`loop-run-log.md`	append-only, one line per beat
`.claude/skills/loop-task/SKILL.md`	the procedure (Claude Code)
`.claude/agents/loop-verifier.md`	the read-only reviewer — this literally IS the source's "independent adversarial review"
`opencode.json.example` · `skills/loop-task.md`	the OpenCode twin — same skill text

Loop Ready checklist

Provable success condition — the source's own spec: reviewer returns ready with no material findings
Run limit — 5 rounds/ticket (this library's number for the source's own "round cap")
Spine written first, committed empty
Report-only (L1) start — writes a draft packet, never the project's canonical spec
Human gate placed — the source's own: round-cap exit logs the blocker for human decision
One log line per beat, no silent runs
Kill switch — loop-pause-all in loop-budget.md

How this relates to the course

This kit is another independently-designed instance of Step 11's maker–checker split, with a detail worth calling out: the source requires the reviewer to "check that prior findings remain resolved," not just catch new ones — a regression check on the review process itself, which is a sharper discipline than most maker–checker loops bother with. Its round cap on an unresolved disagreement is the maker–checker-standoff bound: escalate to the human, don't keep iterating past the cap.

Full attribution: resources/sources.md.