codeaf do "" runs a paid job with an empty brief

[santoshkumarradha]

Seen on: dev 837b2b0.

Behaviour

codeaf do "" (one empty argument) starts a real, paid run. It exited 0 with stop:"done" and spent $0.057; the crew invented a goal from the repository (it fixed the failing tests). A missing brief should be refused the way a terminal with no piped text is refused, with nothing spent.

Replication

1. Build dev 837b2b0 (git checkout 837b2b0 && make build, binary bin/codeaf), or install the dev build with curl -fsSL https://agentfield.ai/get/devaf | bash.
2. Use an isolated profile: export HOME=$(mktemp -d), export OPENROUTER_API_KEY, and keep the default model (~deepseek/deepseek-v4-flash-latest, crew on auto).
3. On a busy machine set task.max_load to 0 (/settings, Tasks) so the busy-machine gate does not hold tasks.
4. Make a small repo with a failing test: R=$(mktemp -d) && cd "$R" && git init -q && printf 'package main\n\nfunc main() {}\n' > main.go && printf 'module demo\n\ngo 1.22\n' > go.mod && git add -A && git commit -qm init, then add any failing _test.go file and commit it.
5. Run codeaf do "" --json; echo exit=$?.
6. Read the JSON: stop is done, spend_usd is non-zero, and git status shows edits nobody asked for.

Evidence

• JSON record: "stop":"done", "spend_usd":0.057, exit 0.
• cmd/codeaf/main.go:1101-1112 (readText): with len(args) > 0 it returns strings.TrimSpace(strings.Join(args, " ")), so one empty argument becomes an empty goal with no error. The noGoalGiven check (main.go:1109, :1121, :1126) runs only on the no-argument and piped paths.

Guessed cause

A guess from reading the code, not a confirmed diagnosis. readText never checks the trimmed argument text for emptiness, so an empty brief reaches the crew and the planner fills the gap from the repository.

Acceptance

• e2e: codeaf do "" and codeaf do " " exit non-zero with the same one-line usage as codeaf do with no argument, and no usage row is written (codeaf logs shows no call).
• Unit: readText returns the noGoalGiven error for an all-blank argument list.

Found while writing the public docs; manual text differences are in #1545.

🤖 Generated with Claude Code

[AbirAbbas]

Fixed on dev — checked on 2f33722b2 (2026-09-30): codeaf do "" --json in a fresh repo exits 1 before any model work, prints error: no goal given, and the JSON reads ok:false, stop:"error", calls:0, spend:0. The empty-brief refusal came with #1604 (v0.5.0).