gan-style-harness

GAN-inspired Generator-Evaluator agent harness for building high-quality applications autonomously. Based on Anthropic's March 2026 harness design paper. Use when a feature should be built autonomously through generator and evaluator iteration until it clears a quality bar.

Quality

65%

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 ./skills/gan-style-harness/SKILL.md

Quality

Content

63%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 well-structured and actionable with concrete commands and a clear iterative workflow, but it is overlong for a SKILL.md overview, inlining material that belongs in separate reference files, and it references bundle files that are not actually present.

Suggestions

Move the full Evaluation Rubric, the env-var/eval-mode tables, and the Evolution/Results sections into separate reference files (e.g. EVALUATION.md, CONFIG.md) and link to them one level deep.

Create the referenced bundle files (PLANNER_PROMPT.md, EVALUATOR_PROMPT.md, scripts/gan-harness.sh) or remove the references so the skill is self-consistent.

Trim the ASCII architecture diagram and 'Results: What to Expect' table to reduce tokens, since the core insight is already stated in prose.

Dimension	Reasoning	Score
Conciseness	Mostly efficient but padded in places — the ASCII architecture diagram, the full Evaluation Rubric markdown block, the 'Evolution Across Model Capabilities' narrative, and the 'Results: What to Expect' table restate context Claude largely already knows; could be tightened without losing actionability.	3 / 5
Actionability	Provides concrete, mostly executable guidance — real shell commands with env-var examples, manual `claude -p` invocation steps, and a config table with defaults — with only minor gaps (the shell script path `./scripts/gan-harness.sh` references a file that is not present in the bundle).	4 / 5
Workflow Clarity	The Plan → Generate → Evaluate → iterate loop is clearly sequenced with explicit iteration/repeat guidance and a pass threshold plus max-iterations cap, and Anti-Patterns add error-recovery context; checkpoints are mostly present though explicit 'stop and validate before next phase' gates are implicit rather than enumerated.	4 / 5
Progressive Disclosure	Well-sectioned but monolithic within SKILL.md: it inlines the full evaluation rubric, env-var tables, evolution narrative, and results table that could live in separate reference files, and references bundle files (PLANNER_PROMPT.md, EVALUATOR_PROMPT.md, scripts/gan-harness.sh) that do not exist in the bundle.	3 / 5
	Total	14 / 20 Passed

Description

67%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 clearly states what the harness does and provides an explicit 'Use when' trigger with a concrete quality-bar condition. It is strong on completeness and distinctiveness but loses points on trigger-term quality for lacking natural, varied user phrasing.

Suggestions

Broaden trigger terms with natural user phrasing such as 'build an app from scratch', 'frontend design tasks', 'production-quality web app', and 'full-stack application'.

Drop or de-emphasize the date-specific provenance clause ('Based on Anthropic's March 2026 harness design paper') which adds tokens without aiding triggering.

Add one or two concrete app-type examples in the description to improve specificity and reduce overlap with generic coding skills.

Dimension	Reasoning	Score
Specificity	Lists several specific concrete actions — 'building high-quality applications autonomously' and 'generator and evaluator iteration until it clears a quality bar' — naming the domain and the iterative generation/evaluation mechanism, with minor gaps in coverage (no mention of specific app types like frontend/full-stack).	4 / 5
Completeness	Explicitly answers both what ('GAN-inspired Generator-Evaluator agent harness for building high-quality applications autonomously') and when ('Use when a feature should be built autonomously through generator and evaluator iteration until it clears a quality bar'); the 'when' is present and explicit but could name more concrete triggering scenarios.	4 / 5
Trigger Term Quality	Contains relevant phrases like 'building high-quality applications autonomously' and 'feature should be built autonomously', but lacks the natural conversational terms a user would say (e.g. 'build an app', 'frontend design', 'production quality') and synonyms that would broaden triggering.	3 / 5
Distinctiveness Conflict Risk	The 'GAN-inspired Generator-Evaluator' framing and 'clears a quality bar' trigger carve a fairly distinct niche from generic build/coding skills, with only minor overlap risk against general application-building skills.	4 / 5
	Total	15 / 20 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: affaan-m/ECC
Path: skills/gan-style-harness/SKILL.md
Commit: 06c5e11

Reviewed: 3 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.