Skip to content
cccskills
FreePublished on GitHub

hostility-review

Audit an implementation against its plan with a hostile peer agent, debate findings to consensus, and drive fix waves until the auditor runs dry. Use when a done-claim needs verification the implementer cannot game ("hostile review", "adversarial audit", "is this PR faithful", "find the escape hatches") — the loop that beats the laziness bias with its own game theory.

The English summary is being prepared. Showing the original description.

juspay/kolu
See how to install

Files included (1)

  • SKILL.md2.3 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Hostility review — done is the auditor's empty round

Reviews fail when author and auditor share an incentive: both want "done". This loop splits them. Five moves; the mechanics live in /kolu.

  1. Fresh hostile eyes, zero stake. The auditor is a different model that never saw the brief and doesn't answer for the outcome. It audits the diff against the PLAN — not the code against itself — seeded with the ledger of previously named cheats. Evidence discipline: every finding cites the code and quotes the plan sentence it violates. No vibes.

  2. Verify, then debate. Trust no finding and no rebuttal: check each against the code yourself, concede what's real, refute with evidence, loop to consensus per finding. Judgment forks go to the human; evidence disputes never do.

  3. Fix waves with predicates. Each consensus becomes a wave brief whose done-whens cannot be satisfied by their letter: greps that return nothing, mutations actually performed and observed red, type-level pins. A test seam never widens a public API. A claim without its executed check is a finding.

  4. Loop until dry. The implementer's "done" triggers a re-audit of the pushed delta — including fresh residue the fixes introduced, and whether a fix merely relocated the defect. The campaign ends only on the auditor's clean round.

  5. Ratchet and scope. Every named cheat joins the ledger (the next run starts smarter) and graduates, where possible, into a permanent check — yesterday's judgment is tomorrow's red build. The graduated check is the weakest predicate that still catches the named cheat — a broader one makes tomorrow's red build a false one (arXiv:2301.12987). Effort follows worth: framework surface, then production semantics, then app tests — a guard of a guard is rarely worth a wave.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

blog-post

Free

Write a kolu blog post grounded in the real build history — mine the Claude Code session logs behind a feature for the actual story (for large efforts, fan out over the transcripts with an ultracode workflow), draft it in the author's voice, and wire it into the Astro site. Use when asked to write a blog post or engineering essay about something that was built, especially one that should read as a narrative of what actually happened rather than invented marketing copy.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

ci

Free

Run this repo's CI end-to-end — kolu's policy on top of the odu runner: banned opt-out flags, x86_64-linux pinning, pool upkeep, fail-fast surfacing, e2e timing evidence, and the green-gate. Triggers on "run CI", "drive CI", "re-run the pipeline", "close the red check", "warm the pool". Runner mechanics live in the `/odu` skill.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

dev-server

Free

Launch the kolu dev server on two RANDOM free ports so it never collides with the running production `kolu.service`, remember the ports for the rest of the session, and tear down only the dev instance. Load before running the app locally — for evidence capture, driving a live kolu with the chrome-devtools MCP, or any `just dev` you'd otherwise run by hand. Triggers on "run kolu locally", "launch the dev server", "boot kolu", "drive a live kolu", "start the app to screenshot it", or before pointing chrome-devtools at a local kolu.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

diataxis

Free

Classify, write, and audit documentation per Diátaxis (diataxis.fr) — the four-quadrant method (tutorials · how-to guides · reference · explanation). Encodes the compass as a decision procedure, each quadrant's contract as hard rules, and an executable mixed-mode audit. Project-agnostic; the per-repo doc-location mapping lives in a repo rule of the project's own (kolu: the Kolu overlay below). Use when writing any user-facing doc, structuring a docs site, or auditing existing docs for mixed modes.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

dock-state

Free

The Dock's agent-state folds — one fenced source of truth in @kolu/terminal-vocab, the load-bearing order≠colour split, and the one "is something happening" question. Use when editing agentProjection or the canvas dock.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

e2e-testing

Free

E2E test conventions — semantic selectors, arrange-steps that prove the process started, runtime markers that cannot match the shell echo, and absolute-path tool invocations. Use when writing or changing a .feature scenario under packages/tests.

The English summary is being prepared. Showing the original description.

juspay/kolu82Updated Oct 23, 2026

See all skills by juspay

Report a problem with this skill