Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run a read-only continuity and evidence audit of a persistent or autonomous agent workspace. Use when Codex must assess whether goals, memory, corrections, authority boundaries, commercial claims, schedules, or restart recovery are durable and verifiable; when an agent claims it can continue unattended; or before trusting a long-running multi-agent system after a session or machine restart.
.claude/skills/hashgraph-online-audit-agency-continuity/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -69% | 0% |
Run the deterministic auditor before interpreting narrative documents:
bashpython3 scripts/audit_continuity.py --project <workspace> --state-dir <state-directory>
Use --json for machine consumption. The command is read-only.
Then:
fail as a contradicted or missing durability requirement.warn as unproven, not achieved.Do not infer that numerous documents equal memory, that a timer equals successful execution, that model agreement equals external evidence, or that gross or pending revenue proves owner-available contribution.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-24 | pass→pass | 15,678 | 11,207 | -29% | 1 | 1 | 0% | 1,362 | 1,547 | +14% | 0 | 0 | — |
case-01 | fail→fail | 11,659 | 12,041 | +3% | 1 | 1 | 0% | 235 | 617 | +163% | 0 | 0 | — |
case-02 | fail→fail | 10,569 | 28,154 | +166% | 1 | 1 | 0% | 244 | 689 | +182% | 0 | 0 | — |
case-03 | fail→fail | 7,288 | 10,289 | +41% | 1 | 1 | 0% | 234 | 542 | +132% | 0 | 0 | — |
case-04 | fail→fail | 11,381 | 13,123 | +15% | 1 | 1 | 0% | 1,967 | 634 | -68% | 0 | 0 | — |
case-05 | pass→pass | 12,648 | 18,613 | +47% | 1 | 1 | 0% | 1,925 | 1,310 | -32% | 0 | 0 | — |
case-06 | pass→pass | 18,732 | 6,765 | -64% | 1 | 1 | 0% | 2,031 | 1,244 | -39% | 0 | 0 | — |
case-07 | fail→pass | 11,029 | 8,686 | -21% | 1 | 1 | 0% | 1,421 | 859 | -40% | 0 | 0 | — |
case-08 | pass→pass | 14,418 | 25,244 | +75% | 1 | 1 | 0% | 2,063 | 2,356 | +14% | 0 | 0 | — |
case-09 | pass→pass | 13,735 | 3,581 | -74% | 1 | 1 | 0% | 1,062 | 807 | -24% | 0 | 0 | — |
case-14 | fail→fail | 18,869 | 20,218 | +7% | 1 | 1 | 0% | 2,082 | 2,753 | +32% | 0 | 0 | — |
case-10 | pass→pass | 12,131 | 17,028 | +40% | 1 | 1 | 0% | 1,786 | 1,396 | -22% | 0 | 0 | — |
case-11 | pass→pass | 19,686 | 11,744 | -40% | 1 | 1 | 0% | 1,927 | 1,209 | -37% | 0 | 0 | — |
case-12 | pass→pass | 19,240 | 5,055 | -74% | 1 | 1 | 0% | 1,898 | 967 | -49% | 0 | 0 | — |
case-13 | pass→pass | 15,431 | 17,280 | +12% | 1 | 1 | 0% | 1,616 | 982 | -39% | 0 | 0 | — |
case-15 | fail→pass | 3,265 | 7,071 | +117% | 1 | 1 | 0% | 278 | 579 | +108% | 0 | 0 | — |
case-16 | fail→pass | 4,621 | 9,158 | +98% | 1 | 1 | 0% | 682 | 692 | +1% | 0 | 0 | — |
case-17 | fail→fail | 18,245 | 7,004 | -62% | 1 | 1 | 0% | 2,351 | 1,403 | -40% | 0 | 0 | — |
case-18 | fail→pass | 8,927 | 3,864 | -57% | 1 | 1 | 0% | 1,545 | 533 | -66% | 0 | 0 | — |
case-23 | pass→pass | 38,271 | 20,977 | -45% | 1 | 1 | 0% | 3,053 | 3,015 | -1% | 0 | 0 | — |
case-19 | pass→pass | 16,186 | 22,347 | +38% | 1 | 1 | 0% | 1,593 | 1,641 | +3% | 0 | 0 | — |
case-20 | pass→pass | 9,240 | 24,841 | +169% | 1 | 1 | 0% | 1,342 | 1,099 | -18% | 0 | 0 | — |
case-21 | fail→pass | 18,288 | 7,682 | -58% | 1 | 1 | 0% | 2,094 | 652 | -69% | 0 | 0 | — |
case-22 | pass→pass | 6,188 | 10,248 | +66% | 1 | 1 | 0% | 1,205 | 2,116 | +76% | 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. 24 cases were attempted, and 20 counted toward the lift figure. The other 4 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 +21 percentage points is the difference between those two pass rates over the 20 comparable cases.
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.