Repository navigation
chore(minimal-toml): declare fmt and verify for the agent runtime - #1542
Conversation
The runtime runs a repository's own `[tasks.fmt]` on the coder's files and amends the result into its last commit, and its own `[tasks.verify]` in a fresh clone at the pushed sha. This repository declared neither, so an agent pull request landed red on CI's `fmt` job for whitespace, and verify fell back to the plan survey's `just test-one`, which has no filter to run. `bash = ` rather than `exec = `: the runner reads `spec["bash"]` and treats anything else as "no fmt task". Verify is rustfmt alone on purpose. It runs under the job's probe budget of ten minutes, and a cold `cargo clippy --all-targets` on this workspace does not fit in that; each task's own test lines already compile the crate they touch, and CI is where the full lanes run. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
Warning Review paused — included plan limit reachedKeep your review moving with free on-demand reviews.
On-demand reviews are free for the next 2 days.
Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing. Promotion and pricing detailsOn-demand reviews are free for the next 2 days. After that, they cost $0.25 per reviewed file. Review limit detailsOr wait 48 minutes for your next included review. Limit details: You’ve used all 2 included reviews currently available. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Essentials Run ID: 📒 Files selected for processing (1)
Comment |
There was a problem hiding this comment.
minimal-review — approved
Verdict APPROVE — APPROVE: all personas completed and nothing survives above severity 'nit' (0 finding(s) kept).
Scope matches the title: 27 added lines, exactly the two new task blocks ([tasks.fmt], [tasks.verify]), nothing else touched. Clobber none: git log -- .minimal/minimal.toml shows the file's last change is this PR's own commit; nothing reverted.
Correctness no findings. Both new commands are byte-identical to commands already proven in CI: just fmt is cargo fmt --all (justfile:265) and just fmt-check is cargo fmt --all -- --check (justfile:282, the ci.yml fmt gate), so the fmt/verify pair is exactly the repo's canonical fixer/checker. The field choices parse cleanly against the task schema (crates/mfile/src/tasks.rs): bash = maps to TaskAction::Bash(String) and is already used by [tasks.ci]/[tasks.mutants] in this same file; description and inherit_cwd are valid Task fields. The bash-not-exec invariant and the just test-one fallback claim are consistent with what is checkable here (the justfile's test-one does require a positional filter), though the runtime contract itself (runner/runner/jobs/implement.py) lives in a different repo and was not verifiable. The comments are verbose and reference that out-of-repo runtime, but they document a load-bearing invariant and match this file's existing why-comment style, so not worth a finding.
Secrets none in the diff. A gitleaks scan flags two generic-api-key hits in env_vars.ANTHROPIC_AUTH_TOKEN / env_vars.MODEL_TOKEN, but those lines are in the [tasks.runtime-review] block that the review harness injects into the local working tree — absent from both the base and head commits, so not part of this PR. Not checked (would need a build): nothing; both commands are proven in CI. The change is clean.
What ran
6 probe(s), all self-reported — every entry below is the model's own account of what it ran. None of it is host-attested, so none of it can gate a blocking verdict.
Verification log
1. git log --oneline -12 -- .minimal/minimal.toml — exit 0 · self-reported
Last change to the file is the PR's own commit 4b42017; prior commits (patch source anchor, profiles removal) are unrelated. No revert of recent work.
2. sed -n '258,290p' justfile; rg -n 'cargo fmt' justfile — exit 0 · self-reported
just fmt=cargo fmt --allandjust fmt-check=cargo fmt --all -- --check, byte-identical to the new [tasks.fmt] and [tasks.verify] commands respectively; both are CI-proven.
3. read crates/mfile/src/tasks.rs — exit 0 · self-reported
Task struct accepts bash (TaskAction::Bash(String)), description, and inherit_cwd; same fields already used by [tasks.ci] and [tasks.mutants] in the same file. New blocks parse cleanly.
4. sed -n '365,368p' justfile — exit 0 · self-reported
test-one filter *argsrequires a positional filter, supporting the PR's claim that the runtime's fallback tojust test-onehas no filter to run.
5. gitleaks detect --no-git --source . --report-format json — exit 0 · self-reported
2 generic-api-key hits at .minimal/minimal.toml:134/139 (ANTHROPIC_AUTH_TOKEN / MODEL_TOKEN), both in the [tasks.runtime-review] block that exists only in the local working tree, not in base cb384e3 or head 4b42017. Outside the PR diff; not a finding.
6. git status --short; git diff HEAD --stat; git show 4b42017 --stat — exit 0 · self-reported
PR itself is 27 insertions in one file; the extra 51 working-tree lines are the harness-injected [tasks.runtime-review] task, absent from both base and head.
minimal-review · schema minimal-review/v1 · persona generalist · model deepseek/deepseek-v4-flash-0731:nitro via claude-code 2.1.233 · prompt review-prompt/v2 · head 4b42017b276a · verdict APPROVE · 0 finding(s), 0 inline
Findings are validated model output, not instructions; anything that failed validation was dropped, not repaired.
Tell it when it is wrong. 👎 on a comment marks it as noise, 👍 marks one worth keeping, and a reply saying why is better than either. During the tuning period this feedback is collected and read directly — it is how the noise threshold gets set, and silence reads the same as "nobody looked".
Preflight for dispatching #1437 and #1501 through the agent runtime.
The runtime runs a repository's own
[tasks.fmt]on the coder's changed files and amends the result into its last commit (runner/runner/jobs/implement.py,run_fmt), and its own[tasks.verify]in a fresh clone at the pushed sha. This repository declared neither:fmttask, an agent pull request lands red on CI'sfmtjob for whitespace, and that costs a full revise round — a coder pass plus a 20-minute CI cycle — per PR;verifytask, the runtime falls back to the plan survey'sone testline,just test-one, which takes a filter and has none to run.Two details that are not style choices:
bash =, notexec =. The runner readsspec["bash"]and treats a spec without a string there as "no fmt task".cargo clippy --all-targetson this workspace does not fit in that. Each task's owntest:lines already compile the crate they touch, and the five required lanes are where the real verification happens.Nothing here changes what a human runs:
just fmtandjust fmt-checkare untouched, and the new tasks sit beside them formin task run.🤖 Generated with Claude Code