openshell-policy-advisor

Use when an OpenShell sandbox returns policy_denied, mentions policy.local, or needs a narrow network policy proposal.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

80%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An extremely lean, mostly actionable single-paragraph workflow with concrete endpoints and a clear approval gate, but it lacks section structure, an explicit error-recovery branch, and copy-paste HTTP examples. Splitting it into a quick-start sequence and signaling the referenced policy_advisor.md would lift structure and actionability.

Suggestions

Break the single paragraph into a numbered quick-start sequence (read -> inspect -> submit -> wait -> retry) so the workflow and approval gate are scannable.

Add concrete executable commands or HTTP examples (e.g., a curl for inspecting /v1/policy/current and submitting to /v1/proposals) so the guidance is copy-paste ready.

Add an explicit error-recovery branch for when the wait times out or the proposal is denied, since the workflow gates destructive retries on approval.

Dimension	Reasoning	Score
Conciseness	A single tight sentence with concrete endpoints, a wait timeout, and an explicit retry condition; it assumes Claude's competence and explains nothing it already knows, so every token earns its place.	5 / 5
Actionability	It names exact endpoints, a wait URL with timeout, and a retry flag ('policy_reloaded: true'), which is mostly executable guidance; it stops short of 5 because no copy-paste command or code block shows how to actually inspect or submit, leaving the HTTP method/payload implicit.	4 / 5
Workflow Clarity	The sequence is clearly ordered (read policy_advisor.md -> inspect current policy -> submit proposal -> wait on chunk -> retry only after approval), with an explicit 'only after approval' checkpoint; it is not 5 because there is no error-recovery branch if approval is denied or the wait times out.	4 / 5
Progressive Disclosure	It references a real bundle file path (/etc/openshell/skills/policy_advisor.md) one level deep, but the body is a monolithic single paragraph with no section headers or signaling of the reference, so structure is minimal and the pointer is buried inline.	3 / 5
	Total	16 / 20 Passed

Description

85%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A well-targeted, third-person description with an explicit 'Use when...' clause and concrete niche triggers; it correctly answers both what and when with low conflict risk. Minor room to add a few natural synonyms to push trigger coverage higher.

Dimension	Reasoning	Score
Specificity	Names a concrete, niche action ('needs a narrow network policy proposal') and a specific trigger state (policy_denied, policy.local), but does not enumerate multiple distinct capabilities, so it stops short of comprehensive coverage at 5.	4 / 5
Completeness	It explicitly states 'when' ('Use when an OpenShell sandbox returns policy_denied, mentions policy.local, or needs a narrow network policy proposal') and clearly implies 'what' (proposing narrow network policies), so both what and when are explicit with concrete trigger phrases.	5 / 5
Trigger Term Quality	It includes natural, user-facing phrases ('policy_denied', 'policy.local', 'narrow network policy proposal') that a user debugging an OpenShell denial would plausibly say, but lacks synonyms or common variations, missing the full 5.	4 / 5
Distinctiveness Conflict Risk	The triggers are highly specific to the OpenShell policy subsystem (policy_denied, policy.local) creating a clear niche with minimal risk of firing for unrelated skills.	5 / 5
	Total	18 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: NVIDIA/OpenShell
Path: crates/openshell-supervisor-process/src/skills/policy-advisor/SKILL.md
Commit: 69a05eb

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.