Spec Runner — Testing skill for Claude Code
Agent skill and CLI that runs a bounded task feature by feature: a features.json queue kept outside the repository, progress.txt as memory across context windows, one landed commit per feature.
How to install Spec Runner
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open ptshih/spec-runner and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Spec Runner does
Agent skill and CLI that runs a bounded task feature by feature: a features.json queue kept outside the repository, progress.txt as memory across context windows, one landed commit per feature. For Claude Code, Pi and Codex.
Alternatives in Testing
- Test Runner — Run Python tests with pytest, unittest, or other testing frameworks 23.4k ★
- Crlf — Test for CRLF / HTTP response-splitting and host-header injection — Set-Cookie injection, cache poisoning, res 4.5k ★
- AI Contribution Skills (Engram) — branch-pr: clean branch and PR workflow 2.5k ★
README
spec-runner
**Run a bounded task feature by feature, one landed commit at a time.**
spec-runner is an agent skill with a small CLI. `features.json` holds the ordered queue, `progress.txt` is the memory that outlives a context window or a session, and `init.sh` is the baseline check. The agent takes the next pending feature, writes the failing test, makes it pass, runs the check, lands the commit, records the pass and continues in the same turn until the queue is complete or a feature is blocked. Nothing external pokes it: the [skill](SKILL.md) supplies the loop, the CLI supplies the state and the guardrails. A pass needs a clean tree and records the landing commit; a queue left in a repository root is refused.
The queue lives outside the repository at `$XDG_STATE_HOME/spec-runner//` (default `~/.local/state`), keyed by the main checkout so a project's worktrees share it. `SPEC_RUNNER_DIR` overrides the location, absolute or relative to the repo root. Nothing in the tree needs ignoring.
Install
The skill directory carries the CLI, so installing the skill installs the tool.
npx skills add ptshih/spec-runner # copies into ~/.agents/skills and links every detected harness
pi install git:github.com/ptshih/spec-runner # as a Pi package
claude plugin marketplace add ptshih/spec-runner && claude plugin install spec-runner@spec-runner
Or clone the repository anywhere your harness reads skills from. Requirements: [Bun](https://bun.sh), git and bash. A shell command is optional:
ln -s ~/.agents/skills/spec-runner/scripts/spec-runner.ts ~/.local/bin/spec-runner
Then ask the agent to run spec-runner, or to work feature by feature. The skill defines `spec-runner` as `bun /scripts/spec-runner.ts`.
Commands
| Command | What it does |
|---|---|
spec-runner init [title] |
Scaffold features.json, progress.txt and init.sh in the queue directory |
spec-runner status |
Print the queue path, completio |
Related Skills
Reconcile
Reconcile git history into the spec corpus — draft stub specs / memory notes for work committed outside FORGE
CR Agents
Brute-force AI method for technical work. Personas anchored to experts in their fields run as Claude Code sub-
API Spec
Generate Zalando-compliant OpenAPI 3.1 specs (one per bounded context) and publish to SmartBear SwaggerHub
Skill Context Diet
Claude Code may truncate CLAUDE.md past 40,000 chars. Compact rules first; test bounded context ablation with
Open Meeting
회의실 폴더 신설 (다 봇 협업 4-file 표준 — 00-context / 01-spec / 02-progress / 03-outcome)
Jared Start
Begin work on an issue — move to In Progress, load full context (body, latest Session note, linked plan/spec),
Related Agents
Cascade Runner
Sub-agent that orchestrates the PDF acquisition cascade for a batch of refs. Delegates the actual work to the
Swe2 Runner
Dispatches a bounded implementation task to an external SWE-2 agent (Cognition devin CLI) inside a named workt
Monitor Agent
Continuously monitors system health, agent status, queue metrics, and workflow progress. Provides real-time ob