vaara-governed-tool-call — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited vaara-governed-tool-call (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Capability skills let an agent do things. This skill records what it was allowed to do, on what basis, and what it actually did, and inserts a human where the law requires one. It wraps the vaara package (no server needed):
resolve it before the agent proceeds. The deployer can designate specific calls as always-escalate.
outcome is appended to a tamper-evident, hash-chained audit trail that exports to a signed regulator package and verifies offline.
The decision is Vaara's, governed by thresholds the operator sets; this skill does not invent verdicts. Authenticity of the final package rests on the signing key and the optional RFC 3161 time anchor, not on this skill.
Use it before any tool call that is consequential, irreversible, or touches regulated data, especially alongside capability skills (for example clinical or genomic database skills). Do not use it for read-only, low-stakes calls; gate the actions a human would want to be able to stop.
pip install vaara (the base install covers gate, escalate, record;pip install 'vaara[attestation]' is only needed for some export options).
(--audit-db), one for the review queue (--queue-db).
Run scripts/governed_call.py for each governed action.
python scripts/governed_call.py gate \
--agent-id <agent> --tool <tool_name> \
--params-json '{"...": "..."}' \
--audit-db audit.db --queue-db queue.db \
[--mode strict|balanced|eco|performance] [--require-review]Read the JSON verdict and the exit code:
human. Surface the action_id to the user and wait. Proceed only after step 2 resolves it allow.
policy and why.
--require-review designates this call for mandatory human oversight: it always escalates regardless of score. Use it for the operations the deployer has decided a human must always sign off (this is the Article 14 designation). The risk is still scored and recorded.
A reviewer lists, then resolves:
vaara review list --db queue.db --status pending
vaara review resolve --db queue.db --queue-id <id> \
--reviewer <who> --resolution allow|deny \
--justification "..." --audit-db audit.db--audit-db writes the ESCALATION_RESOLVED record (Article 14(4)(d) evidence) into the same chain. On allow, the agent may now execute the call; on deny, it must not.
python scripts/governed_call.py outcome \
--action-id <from step 1> --agent-id <agent> --tool <tool_name> \
--result-json '{"...": "..."}' [--severity 0.0] --audit-db audit.dbpython scripts/governed_call.py export-jsonl --audit-db audit.db --out trail.jsonl
vaara trail export-article12 --trail trail.jsonl --key <ed25519.pem> --out pack.zip
vaara trail verify --zip pack.zipThe package carries the full lifecycle (request, decision, escalation, resolution, execution, outcome), a human-readable report, and a verified hash chain. Add --anchor-tsa <url> to fold in an RFC 3161 time anchor (Article 19 existence-in-time evidence).
--mode) and the deployer's--require-review designations. It is governance, not ground truth.
signing the export with a real key and anchoring it. The dev key from vaara keygen --dev is for evaluation only.
tool's own correctness.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.