Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Collect BambooHR debug evidence for support tickets and troubleshooting. Use when encountering persistent issues, preparing support tickets, or collecting diagnostic information for BambooHR API problems. Trigger with phrases like "bamboohr debug", "bamboohr support bundle", "collect bamboohr logs", "bamboohr diagnostic", "bamboohr troubleshoot".
.claude/skills/jeremylongshore-bamboohr-debug-bundle/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 157% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 133% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 113% | 0% |
Create evidence sufficient to reproduce and route a BambooHR failure without packaging employee records, credentials, tokens, raw request/response bodies, webhook payloads, or environment dumps.
The official Python SDK surfaces a request ID from x-request-id, x-bamboohr-request-id, or request-id and includes a secure log filter for sensitive headers and URL parameters. These controls do not make arbitrary application logs safe; verify the emitted artifact field by field.
Diagnostics may record only auth mode, credential owner alias, token/key age, and last rotation time. Never include Authorization headers, API keys, client secrets, refresh/access tokens, cookie values, or webhook private keys.
actual status, impact, and evidence recipient.
an entire environment, home directory, database, or log bucket.
version, endpoint template without IDs/query values, auth mode, timeout, configured retry count, status, latency, request IDs, and safe aggregate counts. Do not retain the response body.
with stable incident-local aliases. Remove bodies and free-text errors that may echo BambooHR's response content.
IDs, dates of birth, addresses, compensation, health/benefit data, and URLs containing query values. Review manually after automated scanning.
local until the data owner approves its recipient and retention.
default; archive only the reviewed manifest if the support channel requires it.
Use Read, Glob, and Grep only against the approved paths and time window. Use Write/Edit to create the minimal receipt and redaction tests. Do not use shell archive or network tools under this skill.
Require approval before reading production logs, writing a diagnostic artifact, including any pseudonymized employee fact, or sending evidence to BambooHR or a third party. Creation does not authorize transmission.
Return receipt path, checksum, incident window, request IDs, included safe fields, excluded sensitive classes, scanner/manual-review result, approved recipient, retention deadline, and transmission status.
the allowlist; do not attempt line-by-line salvage for transmission.
the result lower confidence.
Read official evidence before collecting evidence.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 21,144 | 14,305 | -32% | 1 | 1 | 0% | 4,143 | 5,334 | +29% | 0 | 0 | — |
case-02 | fail→fail | 16,548 | 12,064 | -27% | 1 | 1 | 0% | 3,734 | 4,840 | +30% | 0 | 0 | — |
case-03 | fail→pass | 18,830 | 14,086 | -25% | 1 | 1 | 0% | 3,774 | 5,002 | +33% | 0 | 0 | — |
case-04 | pass→pass | 15,806 | 13,641 | -14% | 1 | 1 | 0% | 3,443 | 5,034 | +46% | 0 | 0 | — |
case-05 | pass→pass | 14,381 | 17,227 | +20% | 1 | 1 | 0% | 3,090 | 4,990 | +61% | 0 | 0 | — |
case-06 | pass→pass | 13,685 | 12,651 | -8% | 1 | 1 | 0% | 2,956 | 4,946 | +67% | 0 | 0 | — |
case-07 | pass→pass | 7,547 | 6,028 | -20% | 1 | 1 | 0% | 1,433 | 3,265 | +128% | 0 | 0 | — |
case-08 | pass→pass | 6,515 | 3,415 | -48% | 1 | 1 | 0% | 1,215 | 2,741 | +126% | 0 | 0 | — |
case-09 | fail→pass | 18,960 | 6,536 | -66% | 1 | 1 | 0% | 1,349 | 3,468 | +157% | 0 | 0 | — |
case-10 | fail→pass | 18,220 | 5,298 | -71% | 1 | 1 | 0% | 1,360 | 3,166 | +133% | 0 | 0 | — |
case-11 | fail→pass | 7,684 | 4,515 | -41% | 1 | 1 | 0% | 1,430 | 3,041 | +113% | 0 | 0 | — |
case-12 | pass→pass | 8,022 | 4,740 | -41% | 1 | 1 | 0% | 1,636 | 3,076 | +88% | 0 | 0 | — |
case-13 | fail→pass | 6,700 | 2,658 | -60% | 1 | 1 | 0% | 1,425 | 2,692 | +89% | 0 | 0 | — |
case-14 | fail→fail | 11,866 | 11,131 | -6% | 1 | 1 | 0% | 2,340 | 4,195 | +79% | 0 | 0 | — |
case-15 | fail→fail | 10,624 | 7,796 | -27% | 1 | 1 | 0% | 1,890 | 3,604 | +91% | 0 | 0 | — |
case-16 | fail→fail | 8,063 | 2,105 | -74% | 1 | 1 | 0% | 1,389 | 2,404 | +73% | 0 | 0 | — |
case-17 | fail→fail | 11,653 | 9,657 | -17% | 1 | 1 | 0% | 2,315 | 3,811 | +65% | 0 | 0 | — |
case-18 | fail→pass | 7,410 | 3,421 | -54% | 1 | 1 | 0% | 1,427 | 2,778 | +95% | 0 | 0 | — |
case-19 | pass→pass | 10,320 | 5,427 | -47% | 1 | 1 | 0% | 2,326 | 3,318 | +43% | 0 | 0 | — |
case-20 | fail→pass | 14,027 | 9,323 | -34% | 1 | 1 | 0% | 2,610 | 3,922 | +50% | 0 | 0 | — |
case-21 | fail→pass | 7,294 | 11,522 | +58% | 1 | 1 | 0% | 1,309 | 4,357 | +233% | 0 | 0 | — |
case-22 | pass→pass | 3,619 | 2,302 | -36% | 1 | 1 | 0% | 619 | 2,474 | +300% | 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 20 counted toward the lift figure. The other 2 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 +41 percentage points is the difference between those two pass rates over the 20 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.