Agent plugin for structured software delivery. Each skill can be used independently or chained together into an end-to-end pipeline.
Rough idea → writing-specs → decomposing-specs → executing-plans → Human review
↑ |
| User approves spec |
+--------------------- feedback -----------------------+
For autonomous work via GitHub issues:
GitHub issue → coder-task → classify (bug/feature) → spec → decompose → execute → PR
- writing-specs — Collaborate with the user to produce a design spec with EARS requirements. This is the human checkpoint — the user reviews and approves the spec before anything gets built.
- decomposing-specs — Autonomously converts the spec into a phased, TDD-enforced task list with
[P]parallel markers. Runs plan-reviewer and task-list-reviewer agents to validate coverage and cross-artifact consistency, then hands off immediately. - executing-plans — Orchestrates implementation by dispatching implementer subagents for each task (parallel for
[P]-marked tasks). Reviews code quality with 4 parallel specialized reviewers and test coverage at phase boundaries. Runs CI checks, final full-spec review, and auto-debug escalation if remediation fails. - Human review — The user reviews the finished implementation. Feedback loops back to a new spec or direct fixes.
Each skill produces a file artifact and can be invoked independently:
writing-specs→docs/plans/YYYY-MM-DD-<topic>-design.mddecomposing-specs→docs/plans/YYYY-MM-DD-<topic>-tasks.md
| Skill | Description |
|---|---|
writing-specs |
Collaborate on a design spec with EARS requirements, clarification markers, and brownfield gap analysis |
decomposing-specs |
Break a spec into a phased, TDD-enforced task list with parallel markers and cross-artifact validation |
executing-plans |
Execute a task list with parallel implementer dispatch, 4 specialized code reviewers, test coverage checks, and auto-debug escalation |
coder-task |
End-to-end: GitHub issue → classify bug/feature → spec → tasks → implementation → PR on a fork |
bugfix |
Produce a 3-section bugfix spec (Current/Expected/Unchanged Behavior) — dispatched by coder-task, not invoked directly |
| Agent | Description |
|---|---|
implementer |
Implements a single task with TDD, tests, and verification |
correctness-reviewer |
Verifies plan alignment, logic correctness, completeness, and edge case handling |
design-reviewer |
Reviews code organization, patterns, naming, and reuse opportunities |
security-reviewer |
Reviews code for security vulnerabilities, injection risks, and secrets handling |
test-quality-reviewer |
Reviews test assertion quality, test design, and anti-patterns |
test-coverage-reviewer |
Verifies requirement-to-test mapping at phase boundaries |
task-list-reviewer |
Validates cross-artifact consistency between spec and task list |
spec-reviewer |
Validates implementation against EARS requirements |
plan-reviewer |
Audits a task list for requirement coverage, TDD enforcement, and CI verification |
auto-debugger |
Fresh-context root-cause analysis when remediation is exhausted |
claude plugin add xmtplabs/code-factoryOr install from a local clone:
git clone https://github.com/xmtplabs/code-factory.git
claude plugin add /path/to/code-factorygit clone https://github.com/xmtplabs/code-factory.git
cd code-factory
mkdir -p ~/.agents/skills
ln -s "$(pwd)/skills" ~/.agents/skills/code-factorygit clone https://github.com/xmtplabs/code-factory.gitAdd to your opencode.json:
{
"plugins": [
{
"name": "code-factory",
"path": "/path/to/code-factory"
}
]
}