| name | adaptive-project-harness |
| description | Route AI-assisted project work into the smallest safe workflow: quick task, planned feature, or bounded long-running work with evidence-first completion. |
Adaptive Project Harness
Use this skill when a coding agent needs a lightweight operating frame for project work.
Before Acting
- Read the target project's local
AGENTS.md or equivalent instructions.
- Read the current task file if one exists.
- Identify the smallest safe route:
- quick task;
- planned feature;
- bounded long-running work.
- Define completion evidence before editing.
Route Rules
Quick Task
Use for small, reversible work that can be checked immediately.
Evidence examples:
- tests pass;
- lint passes;
- one file diff is reviewed;
- a report path is produced.
Planned Feature
Use for multi-step work that still fits in one session.
Requirements:
- short plan;
- affected files;
- checks to run;
- residual risks.
Bounded Long-Running Work
Use for repeated review, migration, repair, or large cleanup.
Requirements:
- max iteration count;
- timeout or stop rule;
- checkpoint state;
- evidence per pass;
- explicit stop when evidence no longer supports more changes.
Safety Rules
- Do not run unbounded loops.
- Do not mutate production or target repositories without an explicit apply path.
- Do not treat memory, summaries, worker output, or external reviewer output as source of truth.
- Do not store secrets, cookies, private keys, raw private logs, live database contents, or full private documents.
- Keep edits scoped and preserve unrelated user changes.
Completion Report
Report:
- route used;
- changed files;
- checks run;
- skipped checks and why;
- evidence;
- residual risk.