Feedback Loop

Never run raw cargo build or npm test from the AI. These tools do the same thing but map errors and failures back to named functions in the graph — so the AI knows exactly what to fix.

code_build

Compile the project. Errors are mapped to function names.

code_build()

On failure, automatically runs code_impact on each broken function — shows what depends on what needs fixing. Fix the CRITICAL dependents first.

---

code_test

Run tests. Failures mapped to functions.

code_test()
code_test(filter="payment")   # run only matching tests

---

code_lint

Lint the project. Warnings mapped to functions.

code_lint()

---

code_exec

Run any arbitrary command. 60s default timeout.

code_exec(cmd="docker ps")
code_exec(cmd="curl -s localhost:3000/health", timeout=15)
code_exec(cmd="echo $PATH", timeout=5)

Use for one-off checks that don’t fit the other tools.

---

Standard loop

code_build()          → see what's broken
code_get(name="...")  → read the broken function
# fix it
code_build()          → confirm it compiles
code_test(filter="...") → run related tests
code_lint()           → check for warnings