Index of every page on this site: llms.txt
All skills
qoderai avatar

/harness-skill-creator

@c6240a8
by qoderaiqoderai/better-harness2.5k stars
208
Run it yourself

Use when bootstrapping or tightening the smallest harness-oriented skill from an existing repository, workflow, evaluation corpus, or harness-analysis chain. Trigger for meta skills, reusable harness workflows, Qoder/Codex plugin skill sets, skill blueprints, and harness-analysis-like skills that need evidence-backed references, validation gates, and qodercli forward-tests.

Use this Skill: https://skilld.dev/gh/qoderai/better-harness/harness-skill-creator

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesbootstrap-patterns.md

≈902 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Bootstrap Patterns

Use this after reading the source chain; extract transferable skill shape, not source policy.

Pattern Extraction

Source surface What to extract What to leave behind
Entry SKILL.md Trigger scope, non-goals, first-hop routing Long explanation and product names
References Evidence boundaries, output contracts, validation gates Source-only organization policy
Scripts Stable commands that prove behavior Exploratory debugging commands
Templates Required output fields and reader contract Decorative examples
Tests Parser-safe assertions and failure cases Snapshot noise
Plugin/host files Discovery path and wrapper shape Host-specific implementation detail
Forward-test logs Repeated model failures to guard against Model self-reports without local proof

For a harness-analysis-like source, the reusable shape is:

  1. short entrypoint
  2. first-hop references
  3. deterministic validator
  4. visible output contract
  5. model smoke that is weaker than local validation

Minimum Skill Shape

Start with one canonical skill:

skills/<name>/
  SKILL.md
  references/<one-pattern-file>.md

Add more only when there is a concrete consumer:

  • scripts/ only for repeated deterministic automation.
  • extra references only when the first reference grows into independent topics.
  • .agents/skills/<name>/ only as a wrapper, generated mirror, or host-only skill; use SKILL.md as its entrypoint.
  • agents/openai.yaml only as concise UI metadata, not behavior ownership.
  • validation gates as target-owned commands or portable Node/Python checks, not ad-hoc Unix-only grep/find snippets.

Forward-Test Loop

Run qodercli from a neutral directory with explicit paths. Keep prompts small.

Prompt contract:

Use $harness-skill-creator from <harness-root> to inspect <target-repo>.
Do not write files or emit tool calls. Return a compact JSON object with:
status, skill_name, trigger, files, source_evidence, validation_gates,
rejection_risks. `files` is string[], rooted under skills/<skill_name>/.
If evidence is thin, use status "insufficient-evidence".
Keep it under 40 lines.

Each iteration records target, command, contract pass/fail, line count, failures to convert into skill rules, and local checks run afterward.

Reject a round when the output:

  • omits frontmatter or trigger ownership
  • emits pseudo tool calls, shell probes, or an inspection plan instead of a skill map
  • proposes broad report generation instead of skill creation
  • returns file objects or paths outside skills/<skill_name>/
  • invents unsupported install, network, server, migration, or dependency-change commands
  • defaults to Unix-only search commands instead of portable or target-owned validators
  • copies source product policy as if it were universal
  • claims filesystem or git cleanliness without local verification

Simplicity Test

Before finishing, delete any rule that does not protect one of these outcomes:

  1. the right skill triggers
  2. the generated skill is short
  3. the generated skill has valid frontmatter and links
  4. the evidence boundary is honest
  5. qodercli failure produces a better next rule

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts3mo3 checks · Risk SAFE
  • Gen Agent Trust Hub3mo

    The skill is a meta-development tool designed to bootstrap and generate other AI agent skills by analyzing existing repositories. It executes local validation scripts and uses vendor-specific CLI tools (qodercli). While it functions as intended for a development workflow, it possesses an inherent attack surface for indirect prompt injection as it processes content from external repositories.

  • Socket3mo

    No alerts

  • Snyk3mo

    Risk: LOW · No issues

Signed by skilld at c6240a8. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last week.

Activeupdated 4 months ago

README badge

README badge for qoderai/better-harness/harness-skill-creator