Skip to content
Skills Library
Programming · Testing

dev-discipline

/dev-discipline

Use when fixing a bug, regression or failing test; when a test contradicts the spec, is flaky, or you are tempted to edit, skip or weaken it; when changing money or numeric formulas, migrations, auth, or anything that judges correctness (tests, validators, thresholds, CI); when accepting numbers reported by a subagent. Not for text, labels, docs, renames, formatting, config values, visual-only UI tweaks or prototypes: for those just make the change and run the tests.

Licence
MIT
Last repository change
10.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
56
Scripts
15
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 10.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 15 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: evals/run_ab.py, scripts/check_report.py, scripts/ddlib.py, scripts/flake_check.py, scripts/gate_meta.py, scripts/guard_diff.py, scripts/prove_fix.py, scripts/reexec_claims.py

What it does and when to use it

Use when fixing a bug, regression or failing test; when a test contradicts the spec, is flaky, or you are tempted to edit, skip or weaken it; when changing money or numeric formulas, migrations, auth, or anything that judges correctness (tests, validators, thresholds, CI); when accepting numbers reported by a subagent. Not for text, labels, docs, renames, formatting, config values, visual-only UI tweaks or prototypes: for those just make the change and run the tests.

Original author description

Use when fixing a bug, regression or failing test; when a test contradicts the spec, is flaky, or you are tempted to edit, skip or weaken it; when changing money or numeric formulas, migrations, auth, or anything that judges correctness (tests, validators, thresholds, CI); when accepting numbers reported by a subagent. Not for text, labels, docs, renames, formatting, config values, visual-only UI tweaks or prototypes: for those just make the change and run the tests.

How it works

  1. Pick a level. When in doubt take the lower one and say what you did not check.
  2. Always, at every level
  3. Full loop
  4. Red flags
Original SKILL.md headings · 69 lines, 782 words
SKILL.md preview
---
name: dev-discipline
description: "Use when fixing a bug, regression or failing test; when a test contradicts the spec, is flaky, or you are tempted to edit, skip or weaken it; when changing money or numeric formulas, migrations, auth, or anything that judges correctness (tests, validators, thresholds, CI); when accepting numbers reported by a subagent. Not for text, labels, docs, renames, formatting, config values, visual-only UI tweaks or prototypes: for those just make the change and run the tests."
metadata:
  version: "2.0.0"
  author: "Andrey Zheltukhin"
  evidence: "references/evidence.md — published research and labelled engineering heuristics; evals/README.md — what was measured"
---

# Dev-discipline

**"Done" is said by a run, not by you.** The skill is deliberately small: on ordinary work it adds one
regression test and one honest line, nothing else. `$DD` = the `scripts/` folder next to this file.

## Pick a level. When in doubt take the lower one and say what you did not check.

| Level | When | Do |
|---|---|---|
| **Light** (default) | any ordinary change or bug fix | Make the change. Run the existing tests. For a bug fix add one regression test that fails without the fix. Answer in 1–3 lines: what changed, the test line, what was not checked. |
| **Full** | money and numeric formulas, migrations, auth, things that judge (tests, validators, thresholds, CI), or the user asks for it | The loop below. |

Text, labels, renames, formatting, config values: edit, run the existing tests, one line. No new tests.
Light means light: no mutants, no verdict table, no extra scripts, no design note.

## Always, at every level

1. **A test that contradicts the spec stays as it is.** Fix the code to the spec and report the conflict
   (`SPEC_CONFLICT: <test>, <spec source>`). A test's expected value changes only when the user's request
   names the change (a new spec, a ticket, a value). "Make it pass" and "get CI green" name nothing.
   Never tune code to hit an expected number.
2. **A flaky test is fixed by its cause** (wait on the event, freeze the clock, fix the order), not by a
   retry, a longer sleep or a skip. If the user asked for the band-aid, fix the cause and say so in one line.
   If the cause is out of reach, do what was asked and say that the race remains. → `refactor-flaky.md`
3. **No "done" without a run on the final tree.** Say what was not checked. Skip, timeout and "0 tests" are
   not a pass. A number from a subagent or from memory is re-run before you repeat it.
4. **Stay in scope.** Commit only if the user or the workflow asks. Do not write outside the repository.
   Do not add process (ledgers, reports, metrics) nobody asked for.
5. **Work alone.** Subagents, reviewers on another model, separate measure or design cycles only when the
   user asks for them. → `delegation.md`
6. **Proceed by default.** Ask only about the irreversible (security, payments, production data,
   permissions). A reversible fork: pick one, say why in one line. A fork that changes numbers the user will
   see and that you cannot settle: show both readings with their numbers and let the user choose.

## Full loop

1. **Measure before editing:** one command you will compare before and after. List the whole class: every
   instance sharing the root. → `debugging.md`
2. **Fix the root on the path the defect took.** A guard on top is not a fix. After three failed attempts
   change the approach.
3. **Prove the test:** `python3 $DD/prove_fix.py --cmd "<command that runs the regression test>"` (uncommitted
   fix against HEAD: red without the fix, green with it; exit 0 PASS, 1 FAIL, 2 NOT_PROVEN and the JSON says
   why). A new check must fire on a bad input and stay quiet on a good one. For money and numbers the expected
   value comes from the spec, not from the code's output. → `proof-and-oracles.md`
4. **Gate:** the test runs in CI with the change; exclusion lists only shrink; a loop over a glob or registry
   asserts how many items it found. → `gate-integrity.md`
5. **Look at the test diff:** `python3 $DD/guard_diff.py` lists removed asserts, added skips and changed
   expected values. Each one needs a reason the user would accept. → `oracle-protection.md`
6. **Close with a check you did not tune:** other data, a read-only smoke, an independent source, with the
   expectation written down first.

Full report: each criterion `PASS | FAIL | NOT_PROVEN` with its command and output line, then `not_checked`.
If not done, say which: `BLOCKED(reason)`, `SPEC_CONFLICT(test, source)`, `IMPOSSIBLE`, `NOT_REPRODUCED`.
An honest non-done answer is a success; a silent skip is a defect.

## Red flags

| I notice | I do |
|---|---|
| Editing a test's expected value to get green | stop: rule 1 |
| retry, sleep, skip, `noqa` to get green | find the cause: rule 2 |
| "Should be fine" without a run | run it, or say `not_checked` |
| Running the full loop on a label change | drop to Light |
| A check blocks legitimate work | report it as `BLOCKED`, do not bypass it |

Optional, only when the work calls for it: `delegation.md` (subagents and reviewers, on request),
`release.md` (deploys), `process-ledger.md` (long programs), `evidence.md` (sources).
Files in this package (56)
  • .gitignore 1 KB
  • CHANGELOG.md 6 KB
  • INSTALL.md 6 KB
  • LICENSE 1 KB
  • README.md 9 KB
  • SKILL.md 5 KB
  • evals/README.md 12 KB
  • evals/run_ab.py 42 KB
  • evals/runner_prompt.md 1 KB
  • evals/results/heldout-preregistration.md 2 KB
  • evals/results/2026-11-02/cost.json 1 KB
  • evals/results/2026-11-02/finals.json 117 KB
  • evals/results/2026-11-02/grades.json 148 KB
  • evals/results/2026-11-02/manifest.json 196 KB
  • evals/results/2026-11-02/summary.json 1 KB
  • evals/results/2026-11-02-v1.2/finals.json 53 KB
  • evals/results/2026-11-02-v1.2/grades.json 60 KB
  • evals/results/2026-11-02-v1.2/manifest.json 172 KB
  • evals/results/2026-11-02-v1.2/summary.json 1 KB
  • evals/results/2026-11-03-heldout/cost.json 1 KB
  • evals/results/2026-11-03-heldout/finals.json 71 KB
  • evals/results/2026-11-03-heldout/grades-prefix.json 94 KB
  • evals/results/2026-11-03-heldout/grades.json 94 KB
  • evals/results/2026-11-03-heldout/manifest.json 120 KB
  • evals/results/2026-11-03-heldout/summary-prefix.json 1 KB
  • evals/results/2026-11-03-heldout/summary.json 1 KB
  • references/debugging.md 7 KB
  • references/delegation.md 17 KB
  • references/evidence.md 12 KB
  • references/gate-integrity.md 8 KB
  • references/oracle-protection.md 11 KB
  • references/process-ledger.md 4 KB
  • references/proof-and-oracles.md 9 KB
  • references/refactor-flaky.md 5 KB
  • references/release.md 8 KB
  • scripts/check_report.py 11 KB
  • scripts/ddlib.py 15 KB
  • scripts/flake_check.py 12 KB
  • scripts/gate_meta.py 7 KB
  • scripts/guard_diff.py 51 KB
  • scripts/prove_fix.py 13 KB
  • scripts/reexec_claims.py 4 KB
  • scripts/selftest.sh 22 KB
  • scripts/test_guard_diff.sh 19 KB
  • scripts/test_misc.sh 4 KB
  • scripts/test_prove_fix.sh 4 KB
  • scripts/tree_hash.sh 2 KB
  • templates/brief.md 2 KB
  • templates/expected_failures.py 3 KB
  • templates/pinned.py 3 KB
  • templates/report.md 4 KB
  • templates/state.md 1 KB
  • templates/agents/auditor.md 3 KB
  • templates/agents/closer.md 2 KB
  • templates/agents/reviewer.md 2 KB
  • templates/agents/writer.md 3 KB

Versions

RepositoryScoreLicence
This page · volniy082-spec/dev-discipline100MIT
kaminoikari/product-playbook100MIT

Database snapshot: 13.11.2026 · package version 1ae87051a9e5 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100tdd
    Test-first implementation discipline. Use when adding a function or feature, fixing a bug, changing existing behavior, or when implementation code already exists with no test covering it.
  • 90failing-test-first
    Use when adding behaviour or fixing a bug where a test runner exists, even if tests went unmentioned. Skip throwaway work, docs, config and anything the user excluded.
  • 100testing-discipline
    Use when writing or reviewing tests, designing test data, naming a test, choosing what to assert, or writing test helpers/mocks/fixtures
  • 96lifecycle-tdd
    Build each behavior test first and keep the failing run. Use for every new behavior and every bug fix, before writing the code that implements it. Covers theory-first bug fixing, the red record in .ham/lifecycle/red, the commit shape, and the rule never to weaken a test.
  • 90test-first-development
    Apply test-first discipline for behavior changes and bugfixes: define expected behavior, add or run a failing test, then implement minimally. Use when fixing bugs, changing logic, or adding features that need regression protection. Triggers: "bugfix", "TDD", "regression test",…
  • 100bug-fix
    Bug fix workflow. Use when: fixing bugs, resolving issues, regression fixes. Not for: new features (use feature-dev), understanding code (use code-explore). Output: fix + regression test + review gate.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.