operations-and-audit
Route general medical utilities, privacy-sensitive operations, document and laboratory work, quality audits, safe catalog queries, and installation/export questions with explicit safety boundaries.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- operations-and-audit
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Route general medical utilities, privacy-sensitive operations, document and laboratory work, quality audits, safe catalog queries, and installation/export questions with explicit safety boundaries.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/audit-and-safety.md9.6 KB
- 参考文档references/installation-and-export.md4.9 KB
- 参考文档references/troubleshooting.md6.6 KB
03 SKILL.md 正文
Operations and Audit
Use this route for the operational layer around the medical-research skill catalog: safe selection of general utilities, PHI/privacy handling, local document/figure/lab operations, audit interpretation, deterministic calculations, and installation or export planning. Keep this file as a router; the detailed rules live in the bundled references.
Triage Before Doing Work
- Protect data first. Treat patient, specimen, accession, EHR, lab-report, DICOM, or chart-shaped content as potentially PHI. Do not echo it, read unknown clinical files into context, or execute a query that may return rows. Apply the minimum-necessary and de-identification rules in audit-and-safety.md.
- Identify the operation. Separate a deterministic local utility from a document/figure/lab workflow, a quality audit, a catalog lookup, or an installation/export request. Confirm inputs, output location, dependencies, overwrite policy, and whether external services are involved.
- Select the smallest catalog skill. Query the generated catalog index by canonical skill id, collection, category, capability, or input format when a lookup helper is available. Do not invent a skill, capability, package, result, or compliance status. Use the routing table below for specialist work.
- Validate before execution. Prefer deterministic, local, reversible operations. Use stable output fields, explicit units/formulas, bounded input ranges, and a validation summary. Stop on missing required inputs or an unsupported dependency rather than silently guessing.
- Apply clinical boundaries. A calculator, converter, de-identification helper, document tool, or protocol utility is research support—not a diagnosis, prescription, treatment recommendation, legal certification, or release approval. Include an appropriate disclaimer and require qualified human review where the result could affect care, dosing, privacy, or release.
Route Map
| User need | Route here | Hand off when |
|---|---|---|
| General calculator, unit conversion, date/gestational-age helper, buffer/dilution or other bounded utility | Use the relevant catalog skill after validating units, ranges, formula, and output contract | A study design decision is needed: protocol-design; statistical/model execution: data-analysis |
| PHI prompt, clinical file/database access, text de-identification, DICOM metadata, privacy review | Apply the privacy gate and route to the relevant privacy skill; require manual QA | Evidence retrieval: evidence-insight; analysis of de-identified data: data-analysis |
| PDF, Word, spreadsheet, slide, image/figure, poster, OCR, figure assembly, lab inventory or preparation operation | Select the narrow local document/figure/lab skill and confirm dependencies and file targets | Manuscript prose, claims, citations, or submission adaptation: academic-writing; analysis: data-analysis |
| Structural or research-quality audit of a skill | Apply the two veto gates and audit workflow in audit-and-safety.md | Specialist category scoring or scientific-content review may require the owning sibling route and human reviewer |
| Catalog discovery or task-to-skill selection | Use the root catalog index and its bundled safe query/check helpers, if present; return canonical id, collection, category, fit, requirements, and limits | Detailed execution belongs to the selected catalog skill, not this router |
| OpenClaw installation, compatible-agent export, or destination collision | Read installation-and-export.md; treat installation as an explicit, reference-only operation | Never install implicitly; ask for authorization and target details before side effects |
Use the sibling handoff only after stating what was learned, what remains unverified, and which inputs the next route needs. Do not duplicate a sibling's specialist workflow in this route.
Compact Operating Contract
- Inputs: a task description, selected skill or operation, local files only when their privacy status is known, required parameters, and an explicit output target for writes.
- Outputs: a bounded plan or local result with objective, inputs and assumptions, selected path, validation status, risks/limits, and next checks.
- Safe failure: name the exact missing field, unsupported pair, unavailable
dependency, collision, or privacy uncertainty; provide the minimum safe
recovery step. Label partial work as
PARTIAL. - No hidden side effects: do not download data, call APIs, install packages, overwrite files, expose PHI, or restart gateways without explicit authority and a reviewed target.
For full rules, consult:
- audit-and-safety.md for veto gates, PHI, deterministic utilities, and clinical disclaimers.
- installation-and-export.md for the OpenClaw reference contract and safe export planning.
- troubleshooting.md for bounded recovery of install, dependency, input, CLI/API, document, figure, and lab failures.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).