Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Register a deferred decision in the debt registry. Trigger by judgment, not a marker scan, whenever a future reader would ask "why this way?": an unmade decision, stub, loosened type, bypassed check, swallowed error, a default picked "for now", or a TODO/FIXME/HACK/XXX marker. Trigger immediately whenever you defer work, or when the user invokes $add. Over-register freely; the developer drops with "drop A", "drop A,C", or "drop all".
.claude/skills/hashgraph-online-add/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-13 | ✓→✗ | ▼ Worse | 32% | 0% |
| case-09 | ✓→✗ | ▼ Worse | 70% | 0% |
| case-12 | ✓→✗ | ▼ Worse | 79% | 0% |
| case-14 | ✓→✗ | ▼ Worse | 58% | 0% |
Call the bundled register.py via Bash — it lives in this skill's scripts/ directory, so reference it with the relative path below (Codex resolves it against the skill root). The helper writes the entry under the repo's detected registry dir (default docs/debt/), assigns a short batch letter (A, B, C…), and prints exactly one line: +1 entry: <slug> (<letter>). That stdout IS the user-facing announcement — add no commentary before or after.
bashpython3 scripts/register.py \ --slug <slug> \ --principal <effort, e.g. 2d, 1w, unknown> \ --interest <ongoing cost, e.g. "+30min/incident", unknown> \ --hotspot <path or module, e.g. pricing/engine.ts, unknown> \ --business-capability <e.g. checkout, billing, unknown> \ --payoff-trigger <concrete trigger, or "unknown"> \ --quadrant <reckless-inadvertent|reckless-deliberate|prudent-inadvertent|prudent-deliberate> \ --category <migration|documentation|testing|code_quality|dead_code|code_rot|expertise|release|infrastructure|planning> \ --ai-authored <true|false> <<'EOF' <body: 2-5 sentences — what the debt is, why it exists, observed symptoms> EOF
The helper:
id itself (no date call needed).~/.cache/debt-ops/cache/<repo-hash>/current-turn.txt (override the base with DEBT_OPS_CACHE) so the user can drop by letter.1–4 word kebab-case label of what the debt is. Examples: cancelled-promotion-callback, legacy-auth-shim, unfinished-rate-limiter. Keep it short — the body carries the context.
reckless-inadvertent (didn't know better), reckless-deliberate (knew, did it anyway), prudent-inadvertent (learned afterward), prudent-deliberate (deliberate, with a payoff plan).unknown ages into stale review and that's the point.drop A, drop A,C, drop all — the user types this; a UserPromptSubmit hook deletes the matching entries and surfaces a one-line confirmation. You don't act on those.drop it or drop <slug> — you delete it yourself: rm <registry-dir>/<id>-<slug>.md (the registry dir named in Discipline 3; default docs/debt/). Treat dropping as cheap — over-registering is the intended posture.register.py.payoff_trigger with a guess to seem certain.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 12,287 | 35,069 | +185% | 1 | 1 | 0% | 1,161 | 2,183 | +88% | 0 | 0 | — |
case-01 | fail→fail | 13,414 | 26,023 | +94% | 1 | 1 | 0% | 1,243 | 2,155 | +73% | 0 | 0 | — |
case-13 | pass→fail | 15,484 | 20,842 | +35% | 1 | 1 | 0% | 1,553 | 2,044 | +32% | 0 | 0 | — |
case-03 | fail→fail | 16,611 | 24,606 | +48% | 1 | 1 | 0% | 1,347 | 1,972 | +46% | 0 | 0 | — |
case-04 | fail→fail | 10,781 | 19,329 | +79% | 1 | 1 | 0% | 1,518 | 1,873 | +23% | 0 | 0 | — |
case-05 | fail→fail | 8,154 | 33,875 | +315% | 1 | 1 | 0% | 1,259 | 1,772 | +41% | 0 | 0 | — |
case-06 | fail→fail | 15,032 | 20,978 | +40% | 1 | 1 | 0% | 1,152 | 1,856 | +61% | 0 | 0 | — |
case-07 | fail→fail | 8,583 | 15,984 | +86% | 1 | 1 | 0% | 1,355 | 1,969 | +45% | 0 | 0 | — |
case-08 | fail→fail | 11,018 | 43,893 | +298% | 1 | 1 | 0% | 906 | 2,241 | +147% | 0 | 0 | — |
case-09 | pass→fail | 7,518 | 30,156 | +301% | 1 | 1 | 0% | 1,311 | 2,230 | +70% | 0 | 0 | — |
case-10 | fail→fail | 15,403 | 35,333 | +129% | 1 | 1 | 0% | 1,638 | 1,904 | +16% | 0 | 0 | — |
case-11 | fail→fail | 15,953 | 25,703 | +61% | 1 | 1 | 0% | 1,681 | 1,721 | +2% | 0 | 0 | — |
case-12 | pass→fail | 9,059 | 36,011 | +298% | 1 | 1 | 0% | 1,250 | 2,237 | +79% | 0 | 0 | — |
case-14 | pass→fail | 10,182 | 17,624 | +73% | 1 | 1 | 0% | 1,232 | 1,942 | +58% | 0 | 0 | — |
case-15 | fail→fail | 10,608 | 16,603 | +57% | 1 | 1 | 0% | 1,245 | 2,387 | +92% | 0 | 0 | — |
case-16 | fail→fail | 7,791 | 25,618 | +229% | 1 | 1 | 0% | 281 | 1,224 | +336% | 0 | 0 | — |
case-17 | fail→fail | 10,861 | 10,096 | -7% | 1 | 1 | 0% | 181 | 1,107 | +512% | 0 | 0 | — |
case-18 | fail→fail | 23,935 | 19,853 | -17% | 1 | 1 | 0% | 1,283 | 1,968 | +53% | 0 | 0 | — |
case-19 | fail→pass | 14,560 | 21,213 | +46% | 1 | 1 | 0% | 1,362 | 2,354 | +73% | 0 | 0 | — |
case-20 | fail→fail | 16,078 | 33,821 | +110% | 1 | 1 | 0% | 1,198 | 1,104 | -8% | 0 | 0 | — |
case-21 | pass→fail | 13,177 | 22,662 | +72% | 1 | 1 | 0% | 1,271 | 1,278 | +1% | 0 | 0 | — |
case-22 | pass→fail | 131,695 | 19,022 | -86% | 1 | 1 | 0% | 3,640 | 1,351 | -63% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 1 counted toward the lift figure. The other 21 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -23 percentage points is the difference between those two pass rates over the 1 comparable cases. 6 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.