harnesslisted
Install: claude install-skill suntao2yl/claude-skill-harness
# Harness
Use one entry point, one ledger, one execution loop, and one completion gate.
Do not expose separate legacy, ledger, or PDCA modes to the user.
Treat the directory containing this `SKILL.md` as `<skill-dir>`. Use:
```text
<skill-dir>/scripts/harness_ledger.py
```
## Merge the responsibilities
Retain the useful responsibilities of the original lifecycle harness without
restoring a second seven-phase state machine:
- Fold necessary discovery, design, architecture, scope shaping, and risk
analysis into Plan. Do only what the approved contract requires.
- Map implementation and other authorized changes to Do.
- Map test, review, release-readiness evidence, and acceptance verification to
an independent Check.
- Map advance, revise, completion, blocking, and durable handoff to deterministic
Act plus the ledger checkpoint.
Scale the Plan artifact to the work. A small fix does not require ceremonial
design or architecture documents; a product-scale change must cover those
decisions before Do. The immutable acceptance contract, not a fixed phase count,
defines the required depth.
The ledger is durable control state, not another work mode. Schema version 1 is
only an approved-contract bootstrap or legacy import. Active delivery runs in
schema version 2 through Plan, Do, Check, and Act.
PDCA execution is Codex-only. Another host may inspect and maintain the durable
contract and handoff state, but must not claim it ran the merged agent loop.
## Shared invariants