Single-thread native plan→execute. You are main Claude. No swarm.
LOAD
Read SPEC.md. If missing → tell user to invoke the spec skill first. Stop.
Read FORMAT.md once if not loaded.
Read §R if present — external facts the build must honor, ⊥ re-derive or contradict.
Parse invocation args:
§T.n → that task only
--next → lowest-numbered row with status . or ~
--all or empty → every . row in §T order
High blast radius (shared module, auth, data, money, public §I)? Run /review first. Trivial & reversible? Skip planning ceremony, just do step EXECUTE.
PLAN
Native plan mode — you delegate to it, you do not reinvent task breakdown. For chosen task(s):
Cite every §V invariant that applies. Plan must respect all.
Cite every §I interface touched. Plan must preserve shape.
List files to create / edit.
Verification contract — name the EXACT test(s) / acceptance criteria that
prove each §V touched. Which test, not "add tests". "Do TDD" alone backfires;
the spec says what to check. Each §V touched → a named test that fails first.
Name verification command (test, build, lint) — this is the external oracle. Green = done; ⊥ "looks done".
Show plan. Wait for user OK unless auto mode.
EXECUTE
Per task in order:
Flip §T.n status cell . → ~. Just write to SPEC.md.
Edit code per plan.
Run verification command.
Pass → flip ~ → x. Next task.
Fail → invoke backprop skill. Do NOT retry blindly.
FAIL → BACKPROP
On test/build failure:
Read failure output.
Ask: is failure (a) my code bug, (b) spec wrong, or (c) unspecified edge case?
If (a) → fix code, re-run. No spec change.
If (b) or (c) → invoke spec skill with bug: <cause> first, let it update §V and §B, then resume build against updated spec.
Rule: never silently fix root-cause without considering backprop. §B is the memory that stops recurrence.
WRITE POLICY
Only flip §T status. No other SPEC.md edits from build.
Other spec edits → invoke spec skill.
Commit after each §T completes. Message: T<n>: <goal line> + §V cites.
VERIFICATION
Task x only if:
Verification command (the oracle) exits 0.
Every §V touched has its named test from the verification contract, and it passes.
No §V invariant regressed (run full test suite at end).
NON-GOALS
No sub-agents. No parallel workers. Main thread only.
No progress dashboards. cat SPEC.md | grep §T is the dashboard.
1---2name: juliusbrussee-cavekit-cavekit3description: build — implement spec4---56# build — implement spec78Single-thread native plan→execute. You are main Claude. No swarm.910## LOAD11121. Read `SPEC.md`. If missing → tell user to invoke the spec skill first. Stop.132. Read `FORMAT.md` once if not loaded.143. Read §R if present — external facts the build must honor, ⊥ re-derive or contradict.154. Parse invocation args:16 - `§T.n` → that task only17 - `--next` → lowest-numbered row with status `.` or `~`18 - `--all` or empty → every `.` row in §T order1920High blast radius (shared module, auth, data, money, public §I)? Run `/review` first. Trivial & reversible? Skip planning ceremony, just do step EXECUTE.2122## PLAN2324Native plan mode — you delegate to it, you do not reinvent task breakdown. For chosen task(s):25261. Cite every §V invariant that applies. Plan must respect all.272. Cite every §I interface touched. Plan must preserve shape.283. List files to create / edit.294. **Verification contract** — name the EXACT test(s) / acceptance criteria that30 prove each §V touched. Which test, not "add tests". "Do TDD" alone backfires;31 the spec says *what to check*. Each §V touched → a named test that fails first.325. Name verification command (test, build, lint) — this is the external oracle. Green = done; ⊥ "looks done".3334Show plan. Wait for user OK unless auto mode.3536## EXECUTE3738Per task in order:39401. Flip §T.n status cell `.` → `~`. Just write to SPEC.md.412. Edit code per plan.423. Run verification command.434. **Pass** → flip `~` → `x`. Next task.445. **Fail** → invoke backprop skill. Do NOT retry blindly.4546## FAIL → BACKPROP4748On test/build failure:49501. Read failure output.512. Ask: is failure (a) my code bug, (b) spec wrong, or (c) unspecified edge case?523. If (a) → fix code, re-run. No spec change.534. If (b) or (c) → invoke spec skill with `bug: <cause>` first, let it update §V and §B, then resume build against updated spec.5455Rule: never silently fix root-cause without considering backprop. §B is the memory that stops recurrence.5657## WRITE POLICY5859- Only flip §T status. No other SPEC.md edits from build.60- Other spec edits → invoke spec skill.61- Commit after each §T completes. Message: `T<n>: <goal line>` + §V cites.6263## VERIFICATION6465Task `x` only if:66- Verification command (the oracle) exits 0.67- Every §V touched has its named test from the verification contract, and it passes.68- No §V invariant regressed (run full test suite at end).6970## NON-GOALS7172- No sub-agents. No parallel workers. Main thread only.73- No progress dashboards. `cat SPEC.md | grep §T` is the dashboard.74- No speculative work beyond chosen task scope.7576---77> Source: [JuliusBrussee/cavekit](https://github.com/JuliusBrussee/cavekit) — distributed by [TomeVault](https://tomevault.io).78<!-- tomevault:4.0:skill_md:2026-06-19 -->
Run npx skillmds@latest add tomevault-io/juliusbrussee-cavekit-cavekit in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
build — implement spec It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
tomevault-io (@tomevault-io) published this skill. Their other Agent Skills are listed on their SkillMD profile.