Repository navigation
Pre-public leak sweep of comments (#30) - #63
Conversation
|
Review: merge after fixes This is not a comment-only change, and the gate changes need another pass. Major
Behaviour changes found (86 non-comment hunks in 36 files; almost all harmless wording or renamed test descriptions): Minor: comments that lost their meaning (9 of 40 sampled)
No new private material found, no test deleted or weakened, no catastrophic regex. |
768404f to
a534c9f
Compare
|
Third review addressed in a534c9f (branch squashed to one commit so the sweep's own intermediate blobs do not sit outside the history cut-off).
Checks: typecheck, boundaries, leaks, history, licence, dashes, lib-version and the gate self-test pass; deck tests 2227 pass; base has the same 4 failures as main. Generated by Claude Code |
e94576b to
df75826
Compare
|
Re-review of df75826: merge after fixes. Much closer. The damaged comments are all repaired, the Major
Minor
Nit
No new private material found. |
Rewrite private context in code comments and doc strings across packages/deck and packages/base, keeping the engineering reasons. Add bounded, attribution-scoped gate rules with self-tests, and a legacy cut-off so the history scan still checks every new blob. Signed-off-by: Claude <noreply@anthropic.com>
df75826 to
af72135
Compare
|
Re-review of af72135: good to merge. The first-name rule is gone, so ordinary sentences no longer trip the gate, and the review-tool rule now needs a change-request reference after the word. The cut-off is pinned in the self-test, the three comments that still read as one deployment (activity, instance, batch-durability) are now generic, and the two nits are fixed. Checks green. One accepted trade-off: with the name rule removed, a real person's first name in a comment is no longer caught by the gate, so that stays a human-review item. |
|
Merge note: this PR conflicts with #62 (overlapping edits in |
Resolve the inbox-push.test.sh conflict keeping both PR 62's byte-escape fixtures and this branch's comment repairs. Note the merge in the SPEC. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Uyne7gFXbMnaS8LJMbfDYR Signed-off-by: Claude <noreply@anthropic.com>
Requested by Simon · project thread
Fixes #30
No real secret or credential was found. Test fixtures that look key-shaped are deliberately fake or split.
What was wrong
Comments and doc strings in
packages/deckandpackages/basestill carried private context the pattern gate cannot see: named people credited as the source of a requirement, internal review process, internal tracker ids, dated incident stories, verbatim quotes of requirements, and provenance notes about a private deployment.What changed
Read-only reviewers read every file; fixes were applied by comment-aware scripts (code untouched), hand edits for narratives, and a final line-by-line repair of every changed comment against its base. The engineering reason was kept in every case; only attribution and context went.
packages/deckapp, lib, test and e2e filestmux.server.ts,activity.server.ts,instance.server.ts,batch-durability.tsinbox,inbox-push,pane-idle,stuck-pane-sweep.sh,activity-sweep.shinbox-push.test.sh,pane-idle.test.shdocs/tooling-review.mdIntended non-comment edits (shipped strings that carried private context), all harmless:
voice-relay.ts,.mjs). Real runtime change, no test either way.avatars.server.ts,inbox.server.ts,queue.server.ts,safe-fs.server.ts,start-production.mjs, androot.tsxguidance no longer carry ids or internal agent names.cron.server.ts(~238) lost an id.projects.server.ts(~496) and placeholder inroutes/projects.tsx(~328) now usePROJ-123.instance.test.ts,projects.test.tsand a few fixtures, with asserted values updated to match. No test deleted, skipped or weakened.Gate (
scripts/leak-gate.mjs, now 33 rules): three new rules with split fragments, self-test samples and negative samples:review-tool-citation(a finding word or tool name followed by a change-request number),internal-reviewer-fold-citation,internal-tracker-prefix-in-comment. They arehistoryExemptonly for legacy blobs, meaning blobs reachable from the pinned commite398eb22; every later blob is scanned with every rule. A missing cut-off fails loudly, the constant is pinned by the self-test, and the self-test proves the bound in a throwaway repo. The tracker-id rule sees lines that start with a comment marker, so an id on an unmarked continuation line of a multi-line JSX comment is a known blind spot (documented in the SPEC).SPEC:
.specs/30-leak 8000 -sweep-comments/SPEC.md, including every review round and how it was addressed.How verified
pnpm typecheck,check:boundaries,check:leaks,check:leaks:history,check:licence,check:lib-version,packages/deck check:dashes: pass.packages/decktests: 2227 passed, 11 skipped (committed tree).node scripts/leak-gate.mjs --self-test: all 33 rules fire; negative samples hold.What could not be verified
check:leaks:messagesis red (23 findings) on commits already on main; none from this branch.packages/basetests: 4 fail here, the same set as on main.check:leaks:historyis clean for main plus this PR; a clone with unmerged remote branches reports their pre-sweep blobs.Follow-ups (not done, owner call)
instancecode keys cron admin on hardcoded lowercase user ids, andprojects.tsxhas one hardcoded second-user option.🤖 Generated with Claude Code
https://claude.ai/code/session_01QYLVaEQcKynDndX32z2sFE