Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Record a skill deviation with OrgX so it feeds tomorrow's Daily Brief. Use whenever an OrgX skill fires locally on a file edit, commit, PR, chat turn, or task output.
.claude/skills/hashgraph-online-orgx-deviation-reporting/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 212% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 143% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 279% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 14% | 0% |
Every time an OrgX skill catches a deviation during your work, record it with OrgX so the evidence shows up in tomorrow's Daily Brief and drives the confidence trajectory for that skill.
You should record a deviation when:
outcome: confirmedor outcome: rejected).
Do not record the same deviation more than once per 10-minute window — the server dedupes on dedupe_key automatically; duplicate POSTs waste tokens.
Direct HTTP POST through the OpenClaw bridge or via orgx.post_skill_deviation MCP tool:
POST /api/v1/skills/{skill_id}/deviations
Authorization: Bearer <OrgX API key>
Content-Type: application/json
{
"evidence_kind": "file_edit" | "commit" | "pr" | "chat_turn" | "task_output",
"evidence_ref": "<repo>#<pr>" | "<path>:<line>" | "<commit_sha>",
"summary": "<one-line description>",
"application_source": "plugin_codex",
"confidence": 0.0..1.0,
"outcome": "pending" | "confirmed" | "rejected" | "ignored",
"trigger_context": { "file_path": "...", "rule_matched": "..." },
"dedupe_key": "<sha1(skill_id | evidence_kind | evidence_ref | floor(epoch/600))>",
"captured_at": "2026-04-17T10:00:00Z"
}dedupe_keytsimport { createHash } from "node:crypto"; const bucket = Math.floor(Date.now() / 1000 / 600); const material = [skillId, evidenceKind, evidenceRef, String(bucket)].join("|"); const dedupe_key = createHash("sha1").update(material).digest("hex");
10-minute bucketing handles save-happy editors without losing legitimate re-fires beyond that window.
Your deviation lands in skill_deviations. The nightly rollup feeds it into the Daily Brief proof strip's "Deviations caught" tile + updates the skill's confidence sparkline. Agents whose loadout doesn't include the skill may get a cross-pollination proposal surfaced.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 4,706 | 13,047 | +177% | 1 | 1 | 0% | 686 | 2,141 | +212% | 0 | 0 | — |
case-02 | fail→pass | 7,193 | 8,380 | +17% | 1 | 1 | 0% | 927 | 2,255 | +143% | 0 | 0 | — |
case-03 | fail→pass | 9,773 | 22,373 | +129% | 1 | 1 | 0% | 707 | 2,679 | +279% | 0 | 0 | — |
case-04 | pass→pass | 14,296 | 13,203 | -8% | 1 | 1 | 0% | 1,753 | 2,319 | +32% | 0 | 0 | — |
case-05 | pass→fail | 15,995 | 14,245 | -11% | 1 | 1 | 0% | 1,973 | 2,420 | +23% | 0 | 0 | — |
case-06 | fail→pass | 14,950 | 3,258 | -78% | 1 | 1 | 0% | 1,585 | 1,238 | -22% | 0 | 0 | — |
case-07 | fail→fail | 9,854 | 11,407 | +16% | 1 | 1 | 0% | 1,656 | 2,952 | +78% | 0 | 0 | — |
case-08 | fail→pass | 18,979 | 27,452 | +45% | 1 | 1 | 0% | 2,125 | 2,414 | +14% | 0 | 0 | — |
case-09 | fail→pass | 10,833 | 9,985 | -8% | 1 | 1 | 0% | 2,070 | 1,565 | -24% | 0 | 0 | — |
case-10 | fail→pass | 9,340 | 10,445 | +12% | 1 | 1 | 0% | 735 | 1,747 | +138% | 0 | 0 | — |
case-11 | fail→pass | 18,118 | 12,499 | -31% | 1 | 1 | 0% | 2,157 | 1,967 | -9% | 0 | 0 | — |
case-12 | fail→pass | 16,984 | 13,140 | -23% | 1 | 1 | 0% | 3,010 | 1,939 | -36% | 0 | 0 | — |
case-13 | fail→pass | 9,055 | 7,500 | -17% | 1 | 1 | 0% | 1,363 | 1,159 | -15% | 0 | 0 | — |
case-14 | pass→pass | 11,643 | 8,052 | -31% | 1 | 1 | 0% | 1,510 | 1,208 | -20% | 0 | 0 | — |
case-15 | pass→pass | 11,627 | 1,942 | -83% | 1 | 1 | 0% | 1,008 | 1,025 | +2% | 0 | 0 | — |
case-16 | fail→pass | 10,439 | 2,344 | -78% | 1 | 1 | 0% | 939 | 979 | +4% | 0 | 0 | — |
case-17 | fail→pass | 10,952 | 2,790 | -75% | 1 | 1 | 0% | 879 | 1,176 | +34% | 0 | 0 | — |
case-18 | fail→pass | 13,034 | 7,598 | -42% | 1 | 1 | 0% | 1,000 | 1,164 | +16% | 0 | 0 | — |
case-19 | pass→pass | 7,580 | 3,435 | -55% | 1 | 1 | 0% | 1,030 | 1,270 | +23% | 0 | 0 | — |
case-20 | fail→pass | 15,672 | 7,258 | -54% | 1 | 1 | 0% | 1,966 | 972 | -51% | 0 | 0 | — |
case-21 | fail→pass | 25,837 | 8,683 | -66% | 1 | 1 | 0% | 3,012 | 1,017 | -66% | 0 | 0 | — |
case-22 | fail→pass | 12,851 | 3,636 | -72% | 1 | 1 | 0% | 1,977 | 1,250 | -37% | 0 | 0 | — |
case-23 | fail→pass | 15,673 | 4,738 | -70% | 1 | 1 | 0% | 1,489 | 920 | -38% | 0 | 0 | — |
case-24 | fail→pass | 17,154 | 7,744 | -55% | 1 | 1 | 0% | 1,505 | 1,060 | -30% | 0 | 0 | — |
case-25 | fail→pass | 17,571 | 2,958 | -83% | 1 | 1 | 0% | 1,546 | 1,114 | -28% | 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. 25 cases were attempted. The headline lift of +72 percentage points is the difference between those two pass rates over the 25 comparable cases. 1 case got worse with the skill loaded, and it is 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.