Skip to content

docs(rfc): redaction that explains itself - #13250

Merged
bolichen97 merged 1 commit into
mainfrom
docs/redaction-rfc
Sep 24, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
docs/redaction-rfc

Conversation

@buluoray

Copy link
Copy Markdown
Contributor

Summary

Docs-only. Adds docs/request-for-change/rfc-redaction-explain-and-reveal.md (plus its interactive prototype and the index row). It proposes that a redaction explains itself where it happens, instead of leaving a bare [REDACTED: …] placeholder.

This RFC was split out of #12684. It lands first so the implementation is reviewed against a decision that is already on the base branch, the same way rfc-crewmates-launch.md went in.

The decision to review

A blocked link keeps its full URL, so a wrongly blocked link can be opened. Most blocked links are false positives, for example wiki pages with Chinese titles or review-filter URLs with long queries. A chip that only explains why a link is gone does not give the reader the link back. §3 explains why this is safe and sets four limits:

  • The gate defends against a request that nobody chose. A person who reads the full URL and then clicks Open once has made that choice.
  • The URL is stored in the message meta. Meta is never replayed to the model, which gets only the role and the message text.
  • A URL that carries a credential is not kept, and its chip has no Open.
  • The removers run again when the user clicks. Only the session owner gets the actions. Every retained field has a limit.

The open questions include the overlapping work in #10965 (a marker on redacted rows), #10038 (media delivery in MCP apps), #11501 (a single redaction-pass helper, which needs a merge order), and the false-positive fixes in #9811, #10192 and #9937.

Implementation

Rollout steps 1 and 3 ship together in one follow-up PR. It replaces #12684 and #13192.

Tested

scripts/docs-lint.sh passes, and the comment-history gate passes.

@buluoray
buluoray requested review from a team and dwu96 September 24, 2026 04:38
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 24, 2026
@github-actions

github-actions Bot commented Sep 24, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 3eeec23eb2479f65daa28e28fd36341fedf20626 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 3eeec23

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 3eeec23eb2479f65daa28e28fd36341fedf20626: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 24, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of 3eeec23eb2479f65daa28e28fd36341fedf20626 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Verify the §Step-1 sibling-channel claims (Teams/Webex/WhatsApp "no server-side unfurl") the same way the RFC verifies Slack's edit path — live, via #12699 — before the implementation PR treats them as closed.

Suggestions

[DESIGN-REVIEWED] 3eeec23

@github-actions

github-actions Bot commented Sep 24, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed 3eeec23eb2479f65daa28e28fd36341fedf20626 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 3eeec23

Verdict parsed from the review's SHA-scoped output markers for commit 3eeec23eb2479f65daa28e28fd36341fedf20626.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 3eeec23eb2479f65daa28e28fd36341fedf20626: <one-sentence reason>

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Sep 24, 2026
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running labels Sep 24, 2026
@bolichen97
bolichen97 enabled auto-merge (squash) September 24, 2026 06:51
@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Sep 24, 2026
Propose explaining every redaction in place instead of leaving a bare
placeholder: a lock tag with its source for a removed value, and a blocked-link
chip for a suspicious URL.

The decision recorded here is that a blocked link keeps its full URL in the
message meta so a wrongly blocked link can be opened once. Most blocked links
are false positives, and a chip that only explains why a link is gone does not
give the reader the link back. The request the gate stops is one nobody chose;
a human who reads the full URL and clicks Open once has chosen. A URL carrying
a credential is not kept and offers no Open, the removers run again at the
click, only the session owner gets the actions, and every retained field is
bounded.

The document lands before its implementation so that the implementation PR is
judged against a decision on the base branch.
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Sep 24, 2026
@bolichen97
bolichen97 merged commit 9c87bcf into main Sep 24, 2026
93 checks passed
@bolichen97
bolichen97 deleted the docs/redaction-rfc branch September 24, 2026 16:17
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Sep 24, 2026
@bolichen97

Copy link
Copy Markdown
Contributor

Holding my approval on one thing only, and it is a one-line change rather than a design objection.

The front matter says status: accepted, and by docs/request-for-change/README.md's own vocabulary that means "Design agreed and locked. Nothing built yet." But the body says the opposite about its own maturity — §3 states "This chip cannot be built from what is saved today" and Rollout Step 1 says "nothing here is on main yet except the widget-frame CSP".

That mismatch has a concrete consequence rather than being a wording nit. The First Principles lane reads a base-branch RFC's status as the product-shape decision record, and treats accepted as decided. So merging this as accepted is what will carry the implementation PR #12684 through that gate — which means the decision it records, keep the whole blocked URL (path and query) in message meta on disk and offer Open once, gets made by a document landing rather than by anyone ruling on it. Today's behaviour is the opposite: the link is destroyed and only the site name survives.

Compare the precedent this RFC cites. rfc-crewmates-launch.md, one of the two accepted RFCs on main, opens with "Status: accepted — the product owner's decision record from the CrewMates launch review of 2026-09-22", and its README line names that decision and its date. There is a signed decision event behind it. This one has no decision line, no decision-maker and no date.

What I am asking for: change the front matter to status: draft and I will approve it as an ordinary docs PR — the writing is good and landing it first so the implementation has something to cite is the right sequence. If it should land as accepted, that needs the owner's own decision on the security posture recorded in the body the way the precedent does it, not a status field.

wknight1 pushed a commit to wknight1/KiroCrew that referenced this pull request Oct 5, 2026
Propose explaining every redaction in place instead of leaving a bare
placeholder: a lock tag with its source for a removed value, and a blocked-link
chip for a suspicious URL.

The decision recorded here is that a blocked link keeps its full URL in the
message meta so a wrongly blocked link can be opened once. Most blocked links
are false positives, and a chip that only explains why a link is gone does not
give the reader the link back. The request the gate stops is one nobody chose;
a human who reads the full URL and clicks Open once has chosen. A URL carrying
a credential is not kept and offers no Open, the removers run again at the
click, only the session owner gets the actions, and every retained field is
bounded.

The document lands before its implementation so that the implementation PR is
judged against a decision on the base branch.

Co-authored-by: Ray Xu <rayrayxu@amazon.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants