01 / Audit
Find a signal
Run a single station to investigate secrets, accessibility, dependencies, performance, documentation or other repository risks.
Browse stations →Observe · propose · optimise
Software Factory
A local-first, schedulable fleet for software-development work. Inspect a station, compose a Line, and route findings without surrendering review.
Run focused agents on demand or in CI. Deterministic tools find candidates; isolated AI sessions judge them; the factory routes verified results to the target’s configured destination.
What can it do?
02 / Compose
Run a Factory Line
Sequence related stations with an Andon policy. A genuine failure or critical defect can stop downstream work; results retain an honest status.
Explore Lines →03 / Integrate
Fit your workflow
Use locally, install a deterministic pre-commit hook, schedule a sweep or run the composite action in GitHub Actions.
See installation routes →04 / Act
Keep humans in control
Track findings in local files or the configured Beads tracker. Proposals are not automatic default-branch changes.
Understand destinations →One pipeline, clear boundaries
- Pre-pass Deterministic discovery supplies bounded candidate evidence.
- Engine An agent skill and adapter triage candidates; separate verification can disprove discoveries.
- Sink Fingerprint and dedupe findings, then use file or configured Beads.
Andon is a stop signal, not a success metric.
Lines record a critical defect or environment failure honestly. An incomplete station never becomes a clean result.
Start with one station
Python 3.9+ and an authenticated supported AI CLI are needed for model work. Start with a local file sink; add tracker routing intentionally.
git clone https://github.com/PaulKinlan/agents.git
cd agents
./install.sh
./factory list
./factory run secret-scan --target . --sink fileFull installation steps · Commands and visibility · How it works under the hood