skill
software-factory
Skill for durable agentic software-factory orchestration
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/jddelia/software-factory itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 22, 2026Software Factory
software-factory is a Codex skill for instantiating and operating adaptive
multi-agent software factories. It helps Codex choose a factory mode, seed
project control docs, assign implementation and review batons, preserve a
single-writer policy, record acceptance evidence, and manage stop/resume
workflow for substantial software projects.
Use it when a project needs more than a one-off edit: end-to-end builds, release readiness, recovery from messy worktrees, safe MVP delivery, strict review gates, or coordinated builder/reviewer workflows.
Website and examples: https://softwarefactoryskill.com/
This skill is best used in the Codex app. It is designed around Codex primitives such as delegated worker threads, tool permissions, browser QA, goal/checkpoint handling, and long-running handoffs. In runtimes with fewer orchestration primitives, use the seeded docs and baton protocol as a manual or single-thread factory workflow.
For a more expansive, DB-backed implementation, see Agentic Factory. That companion Codex plugin includes SQLite-backed durable state, a CLI control plane, event history, baton/review/verification records, pause/resume checkpoints, doctor checks, and markdown ledger rendering. Use this skill when you want a portable, docs-first factory protocol; use the plugin when you want durable state and a larger control plane.
Contents
.
├── SKILL.md # Required Codex skill instructions
├── agents/openai.yaml # UI metadata and default invocation
├── references/configuration.md # Mode presets and policy reference
├── references/templates.md # Factory docs, baton, review, stop/resume templates
├── scripts/seed_factory_docs.py # Dependency-free project doc seeder
├── examples/ # Compact generated-output examples
├── tests/ # Standard-library tests
└── .github/workflows/ci.yml # CI for compile and test checks
Install
Install globally for Codex with the skills CLI:
npx skills add jddelia/software-factory --agent codex --global --yes
To preview what the CLI detects before installing:
npx skills add jddelia/software-factory --list
The skills CLI collects anonymous aggregate install telemetry. To opt out:
DISABLE_TELEMETRY=1 npx skills add jddelia/software-factory --agent codex --global --yes
Manual fallback:
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
git clone https://github.com/jddelia/software-factory.git \
"${CODEX_HOME:-$HOME/.codex}/skills/software-factory"
For local development from an existing checkout:
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills/software-factory"
rsync -a --exclude .git ./ "${CODEX_HOME:-$HOME/.codex}/skills/software-factory/"
Restart Codex after installing so the skill metadata is reloaded.
Use
Invoke the skill directly in Codex:
Use $software-factory to instantiate an adaptive software factory for this project.
Useful variations:
Use $software-factory in Safe MVP mode to ship the thinnest real onboarding slice.
Use $software-factory in Release mode to prepare this repo for publication.
Use $software-factory in Recovery mode to reconcile the dirty worktree and stale handoffs.
The skill will infer or confirm the operating mode, then guide Codex through preflight, factory configuration, baton assignment, verification, review, acceptance, and optional cleanup.
Seed Factory Docs
The bundled generator creates the standard project files that a factory uses as its durable source of truth:
AGENTS.mddocs/factory_config.mddocs/review_index.mddocs/codex_factory_protocol.mddocs/handoff_protocol.mddocs/build_ledger.md
Always dry-run first:
python3 scripts/seed_factory_docs.py /path/to/project \
--work-mode balanced \
--dry-run \
--print-kickoff
Create docs for a Safe MVP factory:
python3 scripts/seed_factory_docs.py /path/to/project \
--project-name "Example App" \
--work-mode safe_mvp \
--target-outcome "Ship the core onboarding flow with focused proof"
The script is conservative: it skips existing files unless --force is passed.
It has no third-party dependencies.
See examples/safe-mvp-kickoff.md for a compact example of the generated file list and kickoff packet.
Validate
Run the local checks:
python3 -m py_compile scripts/seed_factory_docs.py
python3 -m unittest discover -s tests
If you have the Codex skill-creator tooling installed, also validate the skill package:
python3 /path/to/skill-creator/scripts/quick_validate.py .
That validator requires PyYAML.
Design Principles
- Keep
SKILL.mdconcise and trigger-focused. - Keep detailed mode tables, templates, and policy reference material in
references/. - Keep bundled scripts deterministic and dependency-light.
- Preserve user changes and one active writer per worktree by default.
- Treat stop/resume, permissions, model routing, tool-call limits, and verification cadence as explicit factory configuration.
- Record skipped checks and residual risk instead of claiming release readiness without evidence.
Contributing
See CONTRIBUTING.md. Please include tests for generator
behavior changes and keep public skill metadata aligned with SKILL.md.
For vulnerability reports or sensitive security concerns, see SECURITY.md.
License
MIT. See LICENSE.