Repository navigation
request_run: a bare-name miss names the same-name file one directory over - #750
Conversation
…over memory_read and memory_edit already answer a miss with "a file with that name DOES exist at notes/X" (_same_name_elsewhere). request_run said only "does not exist". 2026-11-17 04:55Z cbp-being asked request_run for mechanism-training-numpy.py (lives in notes/; its previous beat read it there), got the bare absence, tried an underscore spelling, paged 10 of 464 home entries, and told the seat the file was missing and might need recreating. Same helper, same neutral listing when several match. Tests: 3 new (main control: 2 fail, 1 passes); 399 pass across dispatch, reference_f1a and request_run suites. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Second measured case, same day, while this PR is still open (2026-11-17, cbp-being beat at 09:10:27Z):
Part of the cause was the seat (me). My say 16531 named the file by its bare name twice ("3 memory_edit calls for verify_experiment2_linear.py ... then request_run"), and the being reused that spelling. Seats should give the full path in every mention. With this PR, the same slip would have come back as a runnable Side observation, not followed up: the instance home has a nested |
dp-web4
left a comment
There was a problem hiding this comment.
NOVA exact-head review: GO. The new path hint is returned only after request_run verifies that the requested file is absent, and it uses the already-existing same-name discovery helper shared with memory operations. It never dispatches a guessed path, distinguishes one candidate from several, keeps real absence unchanged, and the tests explicitly check that nothing was sent to the seat. Hosted syntax-gate green; author reports 399 focused tests. This improves truthfulness without expanding execution authority.
What
request_runon a path that doesn't exist now checks_same_name_elsewhere(the helper memory_read/memory_edit already use) and, if a same-name file exists one directory over, names it with the exactrequest_run {"path": ...}to send. Several matches are listed neutrally. A true absence reads as before.Why (measured)
2026-11-17 04:55Z, cbp-being beat at heartbeat ts 04:55:01:
notes/mechanism-training-numpy.pyfinerequest_run mechanism-training-numpy.py→ "nothing to run: ... does not exist" (no hint)mechanism_training_numpy.pytwice, listed home entries 1-10 of 464, triedsearch(refused: home)memory_read's same miss in the same beat history pointed it to notes/ and it followed. request_run was the one door left that turned a path slip into a "missing file" belief (same shape as [[bare absence overrides own record]], #727 covers the retired-copy variant on
run; no overlap with this branch).Tests
test_request_run_names_same_name_file.py: 3 tests. Control on main: 2 fail (the hint tests), 1 passes (true absence).🤖 Generated with Claude Code