yaml: missing "cleanup" rule — shell-embedded teardown verbs (rm -rf <path>, docker rm/down, kill) inside run:/script: blocks uncaptured #2647

Activity

1. squid-protocol

   added

   enhancementNew feature, sensor, or structural signatureNew feature, sensor, or structural signature

   core-engineModifications to the central physics and parsing engineModifications to the central physics and parsing engine

   on Sep 2, 2026
2. squid-protocol

   added a commit that references this issue on Sep 3, 2026

   fix(core-engine): Rosetta batch 3 wave 2 — doc-comment pairing, rule …

   ...

   deca2bf
3. squid-protocol commented on Sep 4, 2026

   

   squid-protocol

   on Sep 4, 2026

   Last edited by squid-protocol

   OwnerAuthor

   More actions

   Already shipped — closing as done. The rule landed in #2681 ("Rosetta batch 3 wave 2 — doc-comment pairing, rule over-claims, three None rules"), which only listed closes #2672, so this stayed open.

   yaml.py's cleanup is no longer None, and the shipped shape is this issue's proposal with one improvement:

   "cleanup": re.compile(
       r"\brm[ \t]+-rf?[ \t]+(?!/(?:[^A-Za-z]|$))\S{1,200}\b"
       r"|\bdocker(?:[ \t]+compose|-compose)?[ \t]+(?:rm|stop|down)\b"
       r"|\bkill[ \t]+(?!-[0-9])\S{1,64}\b",
       re.M | re.I,
   ),

   Two deltas from the illustrative shape here, both tightening it: the docker alternative also accepts the space form docker compose down (the proposal only had docker-compose), and the rm -rf root guard is (?!/(?:[^A-Za-z]|$)) rather than (?!/(?:[ \t]|$)), so rm -rf / followed by anything non-alphabetic is still left to high_risk_execution instead of leaking into cleanup.

   Verified end to end against engine main (538ef8a):

   • Overlap re-check, which this issue asked for, is settled in the rule's own comment: after_script: stays with func_start (no double-count), and the rule keys off the shell verb rather than the block keyword, so teardown inside run:/script: double-counts with branch/io exactly the way this language's documented philosophy already has shell-in-yaml behave. high_risk_execution's root-only rm -rf / and panics_and_aborts' signal-numbered kill -9 are both preserved — test_yaml_strict.py L63/193 still assert them, and the simple-case table pins run: rm -rf /tmp/build-cache positive against run: rm -rf / negative.
   • Corpus pairing: keyword-rosetta's c.yml plants run: 'rm -rf /tmp/build-artifact' and run: 'docker compose down'. verify_language.py yaml → PASS, 76 assertions across 4 files.
   • Band: the cleanup cell is yaml=2 against a corpus median of 2 — +0%, green. No new red cell manufactured.
   • 122 yaml tests pass, including the ReDoS sweep.

   Same caveat as #2646, and it matters more here because the alternation is broader: no real-world file in the golden master exercises this rule. language-crucible's yaml corpus is 8 Ansible task files with no CI workflow, compose file or teardown verb in them, so a regression in this alternation would not be caught by crucible-audit. Filed as squid-protocol/language-crucible#23.
4. squid-protocol

   closed this as completedon Sep 4, 2026
5. squid-protocol

   added this to the Rosetta Program Comparison milestone on Sep 5, 2026