skill-harness

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This tool acts as an umbrella installer and scaffolding agent. It sets up a shared suite of development skills, agent definitions, and project tooling across Claude and Codex environments.

Security Assessment
The automated code scan found no dangerous patterns, hardcoded secrets, or requests for elevated permissions. However, its core function is executing local shell commands to scaffold projects and install third-party dependencies (including external GitHub repositories like `steveyegge/beads` and other `@45ck` packages). While this is expected behavior for a setup utility, running it means trusting the maintainers of those downstream packages. Overall risk is rated as Medium due to the execution of external, unaudited dependencies.

Quality Assessment
The project is licensed under the permissive MIT standard and is actively maintained, with recent pushes occurring today. It features a detailed, professional README that clearly explains its functions and benchmark results. However, community trust and visibility are currently very low. With only 5 GitHub stars, the tool has not yet been broadly reviewed or validated by the wider developer community.

Verdict
Use with caution — the installer itself appears safe, but you should manually review the external third-party packages it downloads before executing it on your machine.
SUMMARY

Umbrella installer and agent harness for the skill-pack suite across Claude and Codex.

README.md

skill-harness

skill-harness logo

skill-harness is the setup repo for the 45ck agent workflow stack. It is primarily an installer and generator: most downstream repos receive copied Claude/Codex agent definitions or project setup changes rather than importing skill-harness as an application runtime.

It does five jobs:

  • installs the shared dependency-repo suite across pack repos, doctrine repos, and single-skill repos
  • installs the shared Claude and Codex agents
  • bootstraps project-level tooling with @45ck/noslop and 45ck/agent-docs
  • optionally bootstraps Beads, enabled by default in project setup
  • hosts embedded packs for suite-local or incubating capabilities

What it can set up

Shared suite

  • remote dependency repos plus embedded local packs under packs/
  • shared skills synced into ~/.claude/skills/ and ~/.agents/skills/
  • supports repos exposed as skills/, packaged .claude/.agents mirrors, or a single root SKILL.md
  • shared Claude agents
  • shared Codex agents

Project tooling

Use the project setup command when you want a repo scaffolded with the 45ck tooling stack:

./skill-harness setup-project --dir path/to/project

That command:

  • auto-detects monorepo roots from workspace markers such as pnpm-workspace.yaml, package.json workspaces, nx.json, turbo.json, lerna.json, and rush.json
  • auto-detects npm, pnpm, yarn, or bun from lockfiles or packageManager
  • defaults to monorepo-root setup when the target directory is inside a detected monorepo
  • creates a package.json in the resolved setup directory if one does not exist yet
  • installs @45ck/noslop and 45ck/agent-docs
  • installs the Beads CLI if it is not already present
  • runs agent-docs init
  • runs noslop init
  • runs bd init
  • runs agent-docs install-gates --quality

Benchmark results

Controlled experiments measured the toolkit (specgraph + noslop + skills) against raw Claude Code. Full data in experiments/RESULTS.md. Current adoption audit notes are in docs/adoption-audit-2026-04-29.md.

Experiment Toolkit Baseline Delta
Small greenfield (×2 runs) 31–32 / 35 19–20 / 35 +12
Large greenfield (3 modules) 32 / 35 19 / 35 +13
Maintenance / handoff 33 / 35 19 / 35 +14
Ambiguous brief 35 / 35 13 / 35 +22

The gap is driven by scope enforcement and traceability — not code quality (functional output was equal in every experiment). The largest signal came from the ambiguous brief: the toolkit forced scope decisions before code was written; the baseline built a 4-class framework for a task that needed a 3-function library. Treat these as controlled workflow results; downstream repos still need active hook, CI, and agent wiring before those practices are enforced.

Install the CLI

Build locally

git clone https://github.com/45ck/skill-harness.git
cd skill-harness
go build -o skill-harness ./cmd/skill-harness

Windows:

git clone https://github.com/45ck/skill-harness.git
cd skill-harness
go build -o skill-harness.exe .\cmd\skill-harness

Use wrapper scripts

bash install.sh
.\install.ps1

Download a release bundle

Release bundles can ship the binary plus the repo files together so Go is not required.

Build them with:

python scripts/build_release.py --version v0.1.0

Main commands

Install the full shared suite

./skill-harness install --all

Install selected agents

./skill-harness install --agents=requirements-analyst,system-modeler,security-reviewer

Install selected packs only

./skill-harness install --packs=business-analysis-skills,documentation-evidence-skills --packs-only

Install doctrine or utility repos only

./skill-harness install --packs=frontier-agent-playbook,repo-branding-skill --packs-only

Use the interactive installer

./skill-harness install --interactive

Set up a project with noslop and agent-docs

./skill-harness setup-project --dir ../my-project

Install only the packages and skip initialization:

./skill-harness setup-project --dir ../my-project --install-only

Monorepo auto mode:

./skill-harness setup-project --dir ../my-monorepo/apps/web

Force workspace-local setup instead of lifting to the monorepo root:

./skill-harness setup-project --dir ../my-monorepo/apps/web --scope workspace

Override package manager if auto-detection is not what you want:

./skill-harness setup-project --dir ../my-monorepo --package-manager pnpm

Skip one tool:

./skill-harness setup-project --dir ../my-project --skip-agent-docs
./skill-harness setup-project --dir ../my-project --skip-noslop
./skill-harness setup-project --dir ../my-project --skip-beads

Validate installed agent dependencies

./skill-harness check --all

Included agents

  • requirements-analyst
  • requirements-analyst-beads
  • ux-researcher
  • system-modeler
  • system-modeler-beads
  • software-architect
  • software-architect-beads
  • web-engineer
  • backend-engineer
  • test-designer
  • test-designer-beads
  • qa-automation-engineer
  • quality-reviewer
  • security-reviewer
  • security-reviewer-beads
  • pentest-reviewer
  • delivery-manager
  • delivery-manager-beads
  • research-writer
  • workflow-engineer

Agent-to-skill mapping lives in docs/agent-loadouts.md.

Included dependency repos

Shared doctrine companion

45ck/frontier-agent-playbook is the suite-wide doctrine repo for frontier-capability priors, agentic thinking, anti-fallback checks, and LLM-first architecture.

Use it in two ways:

  • install the doctrine skills globally with ./skill-harness install --packs=frontier-agent-playbook --packs-only
  • copy AGENTS.md, CLAUDE.md, AGENT_INSTRUCTIONS.md, and llms.txt into a target project when you want repo-local doctrine surfaces

Tooling repos used here

Full toolkit

The standard full toolkit for a new project is specgraph + noslop. The setup-project command installs both automatically. For manual installation steps or to install the matching skill packs (specgraph-skills, noslop-skills), see AGENT_INSTRUCTIONS.md.

For other agents

If another agent needs to install this repo or use it as the setup entrypoint, point it at AGENT_INSTRUCTIONS.md.

Important files

License

MIT

Yorumlar (0)

Sonuc bulunamadi