rlaope/oh-my-hermes / skills/omh-jev-action-check
omh-jev-action-check
Unverified[omh] Risky command screened by Jev: Jev action check before a risky command: secrets, outbound sends, blast radius; can only add a hold. Use when the user says: jev-action-check, jev action check, ask jev if this command is safe, jev risk check.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.2 KB |
| references/full-contract.md | markdown | 3.8 KB |
| references/preset.md | markdown | 1.4 KB |
references/full-contract.md
View on GitHub# Jev Action Check
Load this on-demand contract after the compact `jev-action-check` body identifies the
workflow. The main body owns routing signals, shared rules, the runtime
operation, delegation fallback, and completion/recovery pointers; this file
holds the workflow-specific detail without copying those shared rails.
## Why This Exists
`jev-action-check` exists so a risky command can get an extra, fail-closed screen from typed answers without Jev ever gaining the power to approve anything.
## Do Not Use When
- The user wants approval or permission policy written; use `omh-security-safety-review`.
- The user wants a command explained, prepared, or run; use `omh-terminal`.
## Examples
Good example:
- Prompt: ask jev if this rm -rf command is safe
- Expected behavior: Say the command, working directory, and task leave the machine, call `action_check/v1`, and report hold or no_extra_hold with the rule before anything runs.
- Why: The user addressed Jev about one command.
Bad example:
- Prompt: is this command risky to run?
- Expected behavior: Route to `omh-terminal`; the user did not ask for Jev.
- Why: A risk question without Jev is the ordinary command lane.
## Completion Checklist
- The outcome was one of hold, refuse_recommended, or no_extra_hold, with its rule.
- Nothing was run on the strength of `no_extra_hold` alone.
## Recovery Notes
- If `omh_jev_ask` is not in the tool list, say Jev is unavailable and the screen did not run (a `hold`, rule `not_answered:tool_unavailable`), point at the `plugin_jev_sidekick` line of `omh doctor`, and run the command only if the user says so after that.
- If the tool returns `consent_not_observed`, nothing was sent; offer the ask in one line naming what would be sent and wait for the user to reply `ask jev`.
- If the tool refuses `credential_like_content` because the command itself carries a credential, report a `hold` (the command carries a credential) first; screen a version with the value replaced by a placeholder only if the user asks, and say the screen did not see the real command.
## When to Use
Use before running a command or write the user asked Jev to screen, especially under a permissive approval mode.
Routing signals: `jev-action-check`, `jev action check`, `ask jev if this command is safe`, `jev risk check`
## Workflow Contract
Category: `review`
Phase: `jev-action-check`
Hermes role: `reviewer`
Quality tier: `evidence-gated`
Reasoning demand: `standard`
Quality bar:
- `state` holds the command, the working directory path, and the stated task; the user was told all three leave the machine.
- The outcome and rule are reported before the command runs.
Handoff policy:
Run in Hermes: build the ask, call `omh_jev_ask`, and report the numbers. Nothing is delegated; the partner workflow keeps its own verdict and Jev's answers are one more input to it.
Required inputs:
- the command or write
- the working directory
- the task the user stated
Expected outputs:
- the `action_check/v1` outcome: hold, refuse_recommended, or no_extra_hold
- the rule that fired
Artifact expectations:
- one metadata-only `omh_jev_ask_record/v1` ledger row per ask under `<omh_home>/jev/asks.jsonl`: hashes, counts, status, usage, cost; never the key, `state`, or question text
Safety rules:
- Call `omh_jev_ask` only after the user asked for Jev in this turn; a skill loaded from the index is not a request.
- Before the first ask, tell the user in one line what `state` will carry and that it leaves the machine.
- A non-answer status is never an answer: report it as a `hold`, tell the user the screen did not run, and run the command only if they say so after that.
- The check can only add a hold: `no_extra_hold` means the host's normal approval still applies, never that the command is approved.
- Every non-answer is a hold labelled `not_answered:<status>`, so the user can see it came from a failure and not from Jev.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |