werkrevisor 0.31.4: deploy-rootfs needs two connections and retries a refused one #70
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: mi/werkrevisor#70
Reference in New Issue
Block a user
See the PR document.
🤖 Generated with Claude Code
🟡 werkrevisor YELLOW for
3628420, superseded by the result forb8d51a6🟡 werkrevisor: YELLOW / HUMAN_REVIEW
Change kind:
mixed.Reasons
Findings
Wasted Sleep on Final Failure at
tools/deploy-rootfs:30-40, severity unknown, counted as medium, hint, as a line commentGiven-When-Then Misuse at
docs/prs/2026-10-10-PR#70-deploy-one-connection.md:45-47, severity unknown, counted as mediumhint; Jev: real 0.73, severity 1.1 of 3
In Scenario#70.02, "Given the host refuses a connection" and "When tools/deploy-rootfs gets the refusal" describe the same event. The Given should establish the precondition (e.g., "tools/deploy-rootfs is deploying an archive") and the When should describe the trigger (the host's refusal), not the tool receiving it.
Implementation Details in Scenario at
docs/prs/2026-10-10-PR#70-deploy-one-connection.md:26-32, severity unknown, counted as mediumhint; Jev: real 0.50, severity 1.1 of 3
Scenario#70.01's "Then" clause ("it opens one connection for the upload and one for checksum, switch and cleanup") lists what each connection does, which is an implementation detail. The scenario would be clearer if the Then described the user-observable outcome, such as "the deploy succeeds without the host refusing connections" or simply "the deploy opens only two connections".
Classifier
4 of 6 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 8784, calls 1, tokens in 8504 / out 17971 (reasoning included in out), duration 130.6 s, cost 0.0241 USD (reported), classifier `jev-1.13.0` calls 6, tokens in 4557, cost 0.0002 USD, finding judge `jev-1.13.0` calls 3, tokens in 4248, cost 0.0002 USD, change summary `openrouter/minimax/minimax-m3` via Minimax calls 1, tokens in 2118 / out 1494, cost 0.0024 USD (reported), PR document check `jev-1.13.0` calls 6, tokens in 8325, cost 0.0003 USD, build file judge `jev-1.13.0` calls 3, tokens in 2994, cost 0.0001 USD, phases: path_judge 0.5 s, classifier 0.5 s, change_summary 7.7 s, doc_check 0.5 s, context 0.6 s, model_review 120.2 s, finding_judge 0.3 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
6418abe).Reviewed head `36284201b799b05199890060a1b447916a00f4f7` against base `6418abe71d19665d2c60df8a56c8cfb7db9f020a`, werkrevisor 0.31.3.
werkrevisor: YELLOW / HUMAN_REVIEW for
36284201b799b05199890060a1b447916a00f4f7.The result comment of werkrevisor holds the details.
@@ -29,3 +30,1 @@scp -q "${ssh_opts[@]}" "$archive" "$target:$directory/"remote_sum=$(ssh "${ssh_opts[@]}" "$target" "sha256sum '$directory/$name'" | cut -d' ' -f1)[[ "$local_sum" == "$remote_sum" ]] || { echo "checksum differs on $target" >&2; exit 1; }retrying() {Wasted Sleep on Final Failure, severity unknown, counted as medium (werkrevisor)
hint; Jev: real 0.76, severity 2.2 of 3
The retrying function sleeps
30 * attemptseconds after every failure, including after the 6th (final) attempt. This prints a misleading "retry in 180 s" message and then wastes 180 seconds before the script gives up. The PR document lists only 5 retry intervals (30, 60, 90, 120, 150 seconds), so the code should perform 5 retries (6 attempts total) without sleeping after the last failure.Fixed in
b8d51a6: after the sixth refused attempt the script gives up at once; it waits only between attempts, 30 to 150 s.Review round of the first head: the hint is fixed in
b8d51a6.🟠 werkrevisor ORANGE for
b8d51a6, superseded by the result forcba9ba1🟠 werkrevisor: ORANGE / FINDINGS
Change kind:
mixed.Reasons
Findings
tools/deploy-rootfs:47-50, severity unknown, counted as medium, ORANGE, as a line commentClassifier
4 of 6 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 8817, calls 1, tokens in 8536 / out 18746 (reasoning included in out), duration 107.3 s, cost 0.0250 USD (reported), classifier `jev-1.13.0` calls 6, tokens in 4557, cost 0.0002 USD, finding judge `jev-1.13.0` calls 1, tokens in 1651, cost 0.0001 USD, change summary `openrouter/minimax/minimax-m3` via Minimax calls 1, tokens in 2131 / out 1255, cost 0.0021 USD (reported), PR document check `jev-1.13.0` calls 6, tokens in 7845, cost 0.0003 USD, build file judge `jev-1.13.0` calls 3, tokens in 3007, cost 0.0001 USD, phases: path_judge 0.4 s, classifier 0.5 s, change_summary 14.0 s, doc_check 0.7 s, context 0.2 s, model_review 90.9 s, finding_judge 0.2 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
6418abe).Reviewed head `b8d51a68add51a68db12c37b1921cf781cbcd5d2` against base `6418abe71d19665d2c60df8a56c8cfb7db9f020a`, werkrevisor 0.31.3.
werkrevisor: ORANGE / FINDINGS for
b8d51a68add51a68db12c37b1921cf781cbcd5d2.The result comment of werkrevisor holds the details.
@@ -34,2 +46,2 @@for repo in "$@"; dossh "${ssh_opts[@]}" "$target" bash -s -- "$repo" "$name" "$version" <<'REMOTE'# The script travels as an argument, so that a retry after a refused connection sends it again.script="release_files=\$(cat <<'PY'Cleanup phase runs invalid Python, severity unknown, counted as medium (werkrevisor)
ORANGE; Jev: real 0.91, severity 3.0 of 3
The cleanup phase invokes
python3 -c "$release_files"on the remote, but the local block that buildsrelease_filesuses a quoted heredoc delimiter (<<'PY'). With'PY'the body is taken literally, so the text$(cat "$(dirname "$0")/release_files.py")is embedded verbatim instead of the contents oftools/release_files.py. On the remote,$0is the placeholderdeploy-rootfspassed tobash -c, sodirnamereturns.;release_filesends up holding an unparsable string andpython3 -cexits with a SyntaxError. The upload and per-repository yml edits run before that point and complete, so the deploy reports success while leaving stale archives and old backups on the host. Drop the quotes around thePYdelimiter (or otherwise embed the script's contents locally) so the body is expanded before being sent.Not a problem, no change: the assignment
script="…"is in double quotes, so$(cat "$(dirname "$0")/release_files.py")runs locally and puts the file into the string; only the escaped\$(cat <<'PY'runs on the host. The run against mih09 with this script removed one archive and 44 backups, which onlyrelease_files.pyselects.Review round of
b8d51a68: the ORANGE finding is wrong (answered in its thread); the scenarios are reworded incba9ba1.🟢 werkrevisor: GREEN / PASS
Change kind:
mixed.Reasons
Classifier
6 of 6 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 8648, calls 1, tokens in 8384 / out 19746 (reasoning included in out), duration 437.7 s, cost 0.0242 USD (reported), classifier `jev-1.13.0` calls 6, tokens in 4562, cost 0.0002 USD, change summary `openrouter/minimax/minimax-m3` via Minimax calls 1, tokens in 2131 / out 2088, cost 0.0031 USD (reported), PR document check `jev-1.13.0` calls 6, tokens in 7286, cost 0.0003 USD, build file judge `jev-1.13.0` calls 3, tokens in 3007, cost 0.0001 USD, phases: path_judge 0.2 s, classifier 0.5 s, change_summary 15.7 s, doc_check 0.4 s, context 0.2 s, model_review 420.3 s, config: defaults + host + repository (.werkrevisor.toml at main6418abe).Reviewed head `cba9ba13c1dfc8244e5a2c76d4808741d8eacda9` against base `6418abe71d19665d2c60df8a56c8cfb7db9f020a`, werkrevisor 0.31.3.
werkrevisor: GREEN / PASS for
cba9ba13c1dfc8244e5a2c76d4808741d8eacda9.The result comment of werkrevisor holds the details.