Back to search
claude-skills
MarketplaceWorks with Claude CodeGitHub
Personal Claude Code skills as a plugin marketplace
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/dknathalage--claude-skills)HTML
<a href="/listings/dknathalage--claude-skills"><img src="/api/listings/dknathalage--claude-skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
/plugin marketplace add dknathalage/claude-skills /plugin install claude-skills@claude-skills
Update with `/plugin marketplace update claude-skills`. The small `marketplace.json` exists only because `/plugin` installs through a marketplace; it points at this repo root.
**`build` needs workflows enabled.** The desktop app has them on. For the CLI, set `"enableWorkflows": true` in your Claude Code settings.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 13 lines
- {
- "name": "claude-skills",
- "owner": {
- "name": "dknathalage"
- },
- "plugins": [
- {
- "name": "claude-skills",
- "source": "./",
- "description": "Personal Claude Code skills bundle."
- }
- ]
- }
README.md · README · 73 lines
- # claude-skills
- Personal [Claude Code](https://claude.com/claude-code) skills in one plugin, `claude-skills`. It replaces [superpowers](https://github.com/obra/superpowers). The discipline is the same (spec → plan → build → review), but the build phase runs **parallel Haiku workers in a Claude workflow**, not one strong-model agent per task in sequence.
- ## Install
- ```
- /plugin marketplace add dknathalage/claude-skills
- /plugin install claude-skills@claude-skills
- ```
- Update with `/plugin marketplace update claude-skills`. The small `marketplace.json` exists only because `/plugin` installs through a marketplace; it points at this repo root.
- **`build` needs workflows enabled.** The desktop app has them on. For the CLI, set `"enableWorkflows": true` in your Claude Code settings.
- ## Skills
- | Skill | Use it when |
- |---|---|
- | `using-claude-skills` | Loaded at session start by a hook. Lists the skills and the rule to check them first. |
- | `spec` | A new feature or change isn't fully specified. A short conversation and 2–3 approaches lead to an approved design doc in `docs/specs/`. |
- | `plan` | An approved spec needs a task graph (`docs/plans/*.json`) where each task owns its files. |
- | `build` | A task graph is ready to run. It runs a Claude workflow: parallel Haiku workers, a Haiku check-and-fix loop per task, one integration step and a whole-branch review on the session model. Per-agent timeouts and a build deadline apply. |
- | `test-driven-development` | You are writing any feature or bugfix code. |
- | `systematic-debugging` | There is a bug, failing test or crash. Find the root cause before any fix. |
- | `verification-before-completion` | You are about to say done, fixed or passing. |
- | `code-review` | You are reviewing work outside a build, or acting on review feedback. |
- | `worktrees` | Work needs isolating in a git worktree. |
- | `finish-branch` | Tests pass and the branch needs merging, a PR, keeping or discarding. |
- | `writing-skills` | You are adding or changing a skill in this plugin. |
- All `SKILL.md` files together come to about 6,800 words, against about 22,000 for superpowers.
- ## How `build` works
- ```
- plan.json ──validate──▶ waves ──▶ Workflow
- ├─ T1: implement (Haiku) → check (Haiku) → up to 2 fix rounds
- ├─ T2: implement (Haiku) → check (Haiku) each task starts as soon
- └─ T3 (needs T1): … as its dependencies pass
- ▼
- integrate (Haiku): merge swarm/<topic>/<id> into build/<topic>, run verify_all
- ▼
- review (session model): whole branch against the spec, one summary back
- ```
- Workers edit only the files their task owns. A task that keeps failing, or runs past its time limit, comes back `stuck` for the session to re-plan. It is never retried on a bigger model.
- ## Benchmarks
- Headless A/B runs against superpowers on .NET fixtures, scored by hidden tests and a blind judge. Details are in [`bench/results/`](bench/results/).
- Results on F-medium (8 tasks), median of 3 runs each:
- | Arm | Time | Cost | Hidden tests | Review |
- |---|---|---|---|---|
- | swarm | 401 s | $1.48 | 20/20 | 8 |
- | same tasks one at a time | 570 s | $1.50 | 20/20 | 8 |
- | superpowers | 695 s | $3.88 | 20/20 | 9 |
- Run one session with `bench/run.sh`, or the E1 set with `bench/e1.sh`. `npm test` runs the harness and workflow tests.
- ## Adding or changing a skill
- See the `writing-skills` skill. In short:
- 1. Edit `skills/<name>/SKILL.md`.
- 2. Bump `version` in `.claude-plugin/plugin.json`.
- 3. Run `claude plugin validate .`.
- 4. Commit, push, then run `/plugin marketplace update claude-skills`.
- ## License
- MIT. Several skills are rewritten from superpowers by Jesse Vincent, also MIT; see [NOTICE](NOTICE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.