mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-07 05:46:59 +02:00
nguyenm7andClaude Opus 5
eb8be96ec5
Merging 89 commits of master added 7 connectors, which re-staled the count
claims the Enterpret acceptance test had flagged. Re-measured rather than
edited by hand.
SKILL.md's credential canary said "72 of 72" and "0 of 72". Re-executed both
patterns over the definitions at d56a8a39d6: the naive
/bearer|token|secret|key/i fires on 79 of 79, the value-shaped one on 0 of 79,
and it still matches sk-, ghp_ and xoxb- test strings. The three legitimate
keyPlacement prefixes ("Bearer ", "Basic ", "Token token=") are still exactly
three.
catalog-contract.md's corpus finding said 72 and 73. Re-ran
`node scripts/ingest-app-definitions.mjs --definitions-only` at this head with
no corpus and PAPERCLIP_CONTENT_TEMPLATES unset: "Parsed 0 captures and 0
states; emitted 79 Wave 1 definitions", git status clean. The finding holds at
the new scale.
Both now say the total is not the claim and tell the reader to re-run, which
is what offline-verification.md already did for its own count ("compare the
fidelity OK line, not the number"). That is the durable fix for this finding
class, not another hand-edited number.
Checked and left alone: the 20 hidden slugs and 56 store-visible rows are
still exact. 22 manifest rows carry catalogVisible:false, but two of those are
not in APP_STORE_HIDDEN_SLUGS, so the documented 20/56 are right and the
naive manifest count is the wrong measurement. The CHAT-CONNECTOR-UX.md and
GENERIC-REMOTE-MCP.md line citations all still resolve to the quoted text
after the merge.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>