behavioral-evals

Guidance for creating, running, fixing, and promoting behavioral evaluations. Use when verifying agent decision logic, debugging failures, debugging prompt steering, or adding workspace regression tests.

Quality

68%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./.gemini/skills/behavioral-evals/SKILL.md

Quality

Content

61%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.

The body is an effective, lean progressive-disclosure router that sequences a clear decision tree and checklist and points to detailed references one level deep. Its weaknesses are delegated executable detail (no code or commands in the body), vague in-body verification, an orphaned bundle file (running.md), and a brittle absolute path to evals/README.md.

Suggestions

Add the concrete run command inline in the 'Verify' checklist step (or link running.md) so the body is actionable without jumping to a reference.

Surface references/running.md in the Bundled Resources list so no bundle file is orphaned from the overview.

Replace the absolute file:// path to evals/README.md with a repo-relative link to avoid a brittle, machine-specific reference.

Dimension	Reasoning	Score
Conciseness	The body is a lean routing document that delegates detail to references rather than padding, but a light justification sentence ('They are critical for verifying prompt changes, debugging steerability, and preventing regressions') and linking 'creating.md' four times keep it just below the every-token-earns-its-place anchor 5.	4 / 5
Actionability	It names concrete tokens like 'rig.setBreakpoint()', 'rig.readToolLogs()', and the 'USUALLY_PASSES' / 'ALWAYS_PASSES' policies, but the body contains no executable code or actual commands (e.g., 'Run single tests locally with Vitest' gives no command), leaving key execution details delegated to the references.	3 / 5
Workflow Clarity	The Workflow Decision Tree and Quick Checklist give a clear sequenced flow with a 'Verify' checkpoint, but that verification is vague ('Confirm stability locally') and the explicit feedback loops (fail-first, run 3× across models) live only in the references rather than the body.	3 / 5
Progressive Disclosure	It is a clear overview with well-signaled one-level-deep references to creating.md, fixing.md, and promoting.md, but references/running.md is a bundle file never linked from the body and the 'Single Source of Truth' pointer to evals/README.md uses a brittle absolute file:// path — minor organization gaps that hold it below anchor 5.	4 / 5
	Total	14 / 20 Passed

Description

76%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.

The description is well-structured with an explicit 'Use when…' trigger clause and a clear, fairly comprehensive action list, scoring high on completeness. Its main weaknesses are missing common trigger synonyms (especially 'evals') and a noun-phrase 'Guidance for…' opening rather than verb-led third-person voice.

Suggestions

Lead with third-person action verbs (e.g., 'Creates, runs, fixes, and promotes behavioral evaluations…') instead of the noun-phrase 'Guidance for…' framing.

Add the natural abbreviation 'evals' and synonyms like 'agent tests' to the trigger terms so users who say those phrases match this skill.

Sharpen the generic 'debugging failures' trigger to something more eval-specific to reduce overlap with general debugging skills.

Dimension	Reasoning	Score
Specificity	Names the domain plus many concrete actions — 'creating, running, fixing, and promoting behavioral evaluations' and 'verifying agent decision logic, debugging failures, debugging prompt steering, or adding workspace regression tests' — but the noun-phrase 'Guidance for…' framing is less concrete than verb-led third person, keeping it just below the comprehensive anchor 5.	4 / 5
Completeness	It clearly states what ('Guidance for creating, running, fixing, and promoting behavioral evaluations') and explicitly when ('Use when verifying agent decision logic, debugging failures, debugging prompt steering, or adding workspace regression tests'), matching the anchor for explicit what-and-when with concrete trigger phrases.	5 / 5
Trigger Term Quality	Includes relevant keywords like 'behavioral evaluations', 'debugging failures', and 'workspace regression tests', but misses common variations users naturally say — most notably the abbreviation 'evals' that the skill itself uses throughout — so it sits at 'some relevant keywords but missing common variations'.	3 / 5
Distinctiveness Conflict Risk	The 'behavioral evaluations' / 'agent decision logic' niche is mostly distinct with minimal conflict risk, though the generic trigger 'debugging failures' creates minor overlap with general debugging skills, placing it just below the minimal-conflict anchor 5.	4 / 5
	Total	16 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
relative_links	Relative link issues: 2 suspicious	Warning

	Total	15 / 16 Passed

Repository: google-gemini/gemini-cli
Path: .gemini/skills/behavioral-evals/SKILL.md
Commit: 85aca16

Reviewed: 4 days 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.