| name | harness-diagnostics |
| description | Assess whether an existing repo is ready for Codex-heavy development. Use before large harness work, when a user asks for an audit, roadmap, maturity score, or wants to know which Pax Fullstack Harness skills to apply first. |
Harness Diagnostics
Purpose
Score repo harness maturity and produce a prioritized roadmap before changing broad project structure.
Inspect First
- manifests, README,
AGENTS.md, docs, scripts, CI, env examples, tests, runtime files, security docs, migration files, release docs, and recent diffs
- existing harness assessment or roadmap docs
- validation command output when safe to run
Procedure
-
Inventory the repo.
- Use manifests and files on disk, not assumptions.
- Run
scripts/assess_harness.py from this plugin when available.
-
Score maturity.
- Agent instructions
- Instruction signal-to-noise
- Repo-local knowledge
- Validation commands
- Testing
- Architecture boundaries
- Local runtime
- Observability
- CI/CD
- Security
- Dependency governance
- Migrations/schema
- Release readiness
- Review protocol
- Entropy controls
- Feature state and WIP control
- Session continuity
- Initialization readiness
- Termination evidence
- Process observability
- Review-feedback promotion
- Skill compliance, context budget, and daily-vs-library placement
- Pre-edit investigation evidence for risky changes
- Browser click-path QA for UI workflows
- Research evidence for external API, dependency, security, or integration choices
- Agent failure-loop risks: weak inspection, tool misuse, wrapper no-ops, hidden repair loops, and premature completion
- Trajectory risk artifacts: task trace, policy, advisory risk wrapper, and risk report
- Project-scoped subagents for explicit exploration, review, docs research, browser QA, and trajectory auditing
- Five subsystem maturity: instructions, tools, environment, state, and feedback
- Treat files as evidence only when they are usable: docs should not be empty placeholders, scripts should be executable, wrappers should not be unconditional no-ops, and validation should wire the expected commands.
-
Write outputs.
docs/harness-assessment.md
docs/harness-roadmap.md
-
Prioritize next skills.
- Recommend the smallest sequence of harness skills that closes the highest-risk gaps.
- Separate blockers from nice-to-have polish.
- Recommend
feature-state-machine, session-continuity-harness, initialization-readiness-harness, termination-gatekeeper, sprint-contract-observability, or review-feedback-promoter when control-loop evidence is missing.
- Recommend
skill-compliance-auditor, agent-failure-introspection, pre-edit-investigation-gate, click-path-and-browser-qa, or research-first-change when ECC-derived reliability gaps are found.
- Recommend
trajectory-risk-supervisor or subagent-review-orchestrator when traces, risky transitions, or parallel review/research would improve reliability.
Validation
- Run
scripts/assess_harness.py <repo> --json when possible.
- Confirm findings cite real files or missing files.
- Confirm a high score is backed by
evidence and has no serious warnings; 100/100 means usable artifacts, not just matching filenames.
- Confirm subsystem scores identify weak instructions, tools, environment, state, or feedback layers.
- Confirm roadmap items map to named skills.
Completion Criteria
- The repo has a clear maturity score, evidence, and ordered next steps.
- The assessment avoids speculative claims and does not mutate the repo except the requested report files.