This demo shows the core AgentFit story: agent instructions drift, AgentFit catches the drift, and a focused fix improves the score.
examples/demo/bad contains a deliberately stale AGENTS.md:
@docs/setup.mdis referenced but missing.pnpm lintis documented butpackage.jsonhas nolintscript.- no runnable verification command is available from the instructions.
packages/apihas a package but no nestedAGENTS.md.
Run:
cd examples/demo/bad
npx @kingkyylian/agentfit@latest eval \
--format markdown \
--output ../../reports/demo-before.md \
--json-output ../../reports/demo-before.json \
--tasks 5 || trueResult:
**Score:** 65/100 (D)
4 failed checks found.
Task execution: static dry-run preview; generated tasks were not executed.
Failed checks:
- No nested instruction file found for packages/api.
- Documented command references missing package script "lint".
- No runnable verification command found in instruction files.
- 1 instruction reference is missing or invalid.
examples/demo/fixed fixes the same repo shape:
- adds
docs/setup.md. - adds runnable
test,lint, andbuildscripts. - documents setup and verification commands.
- adds
packages/api/AGENTS.md.
Run:
cd examples/demo/fixed
npx @kingkyylian/agentfit@latest eval \
--format markdown \
--output ../../reports/demo-after.md \
--json-output ../../reports/demo-after.json \
--tasks 5Result:
**Score:** 93/100 (A)
No failed checks.
Failed checks: none.
From the project root:
npx @kingkyylian/agentfit@latest compare \
examples/reports/demo-before.json \
examples/reports/demo-after.json \
--format markdownResult:
AgentFit improved by 28 points: 65/100 (D) -> 93/100 (A).