[AUTOMATED] fix(brief): a probe is one argv and there is no shell — 6 of 26 were lost to this - #443
Merged
Merged
Conversation
…e round's gate A tester report whose probe arrived as a JSON string that will not parse crashed `verify --gate`, so no observation in the round got gated at all. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fails on the unpatched tree with the AttributeError, passes as `unrunnable` with the guard. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… of 26 were lost to this Round 4's gate: 20 admitted, 6 UNRUNNABLE. That is 23% of the round's evidence discarded, and the discarded set includes "Misspelled option is silently accepted", one of the two most interesting findings of the round. The cause is not a bug. Four of the six wrote their probe as `sh -c '<pipeline>'` and the allowlist refused it: ProbeError: probe may not execute 'sh': it runs code passed as an argument, which would make the allowlist meaningless Two more used `timeout`, refused for the same reason. The guard is correct and stays -- probes are LLM-authored and replayed in the main tree outside the sandbox, which is exactly why argv[0] is allowlisted by resolved realpath. The brief was the gap. It showed a worked example whose `cmd` happens to need no shell and never said a shell is impossible, so a tester reaching for `| grep` to check output had no way to know the probe would be thrown away rather than run. Same shape as the `--assert` quoting trap: the tool is right, the instruction is incomplete, and evidence is lost silently. Adds the rule with a translation table, because "no shell" alone would leave testers stuck: `| grep X` is `stdout_matches`, `| grep -v X` is `stdout_absent`, `| jq .count` is a `json` path clause, and `timeout 60` is the `timeout_s` field that already exists. Also says plainly that a weaker probe which RUNS beats a perfect one the gate discards -- the failure here was testers writing better assertions than the format accepts. tools/repipe/smoke.sh 138/138. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Round 4's gate: 20 admitted, 6 UNRUNNABLE. That is 23% of the round's evidence
discarded, and the discarded set includes "Misspelled option is silently accepted",
one of the two most interesting findings of the round.
The cause is not a bug. Four of the six wrote their probe as
sh -c '<pipeline>'andthe allowlist refused it:
ProbeError: probe may not execute 'sh': it runs code passed as an argument,
which would make the allowlist meaningless
Two more used
timeout, refused for the same reason. The guard is correct and stays --probes are LLM-authored and replayed in the main tree outside the sandbox, which is
exactly why argv[0] is allowlisted by resolved realpath.
The brief was the gap. It showed a worked example whose
cmdhappens to need no shelland never said a shell is impossible, so a tester reaching for
| grepto check outputhad no way to know the probe would be thrown away rather than run. Same shape as the
--assertquoting trap: the tool is right, the instruction is incomplete, and evidenceis lost silently.
Adds the rule with a translation table, because "no shell" alone would leave testers
stuck:
| grep Xisstdout_matches,| grep -v Xisstdout_absent,| jq .countis a
jsonpath clause, andtimeout 60is thetimeout_sfield that already exists.Also says plainly that a weaker probe which RUNS beats a perfect one the gate discards --
the failure here was testers writing better assertions than the format accepts.
tools/repipe/smoke.sh 138/138.
Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com
🤖 Generated with Claude Code
https://claude.ai/code/session_01YcFmfZndNjgfqLQVBZCdkY