Skip to content

docs(rfc): redact credentials in user-typed transcript rows at rest - #29685

Closed
rnoack1 wants to merge 1 commit into
mainfrom
rfc/redact-user-rows-at-rest
Closed

rnoack1 wants to merge 1 commit into
mainfrom
rfc/redact-user-rows-at-rest

Conversation

@rnoack1

@rnoack1 rnoack1 commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Record, as a draft RFC, the decision to scrub credentials out of the transcript rows people type, so a maintainer can accept or reject it.

A credential someone pastes into a channel is stored in the transcript as typed. _redact_at_write_boundary returns user rows unscrubbed, and every dashboard reader of that conversation is served the row. #10965 removes that exemption, which reverses a rule the code pins, and its First Principles lane blocks until the decision is on record.

Why it matters

In a shared channel, more than one person reads the same transcript, so a pasted key outlives the moment it was typed. The fix costs something real: the rewrite cannot be undone. That trade is a maintainer's call, and without a record #10965 cannot merge.

Not a goal

What changed (motivation → approach → change)

A new draft RFC, rfc-redact-user-rows-at-rest.md, and its row in the docs/request-for-change/README.md index. The RFC states the problem, the proposed rule (scrub every role at write, plus the title and live-slot copies), the costs accepted, three alternatives and three open questions.

Backwards compatibility

Compatible: docs only; no code, config or test changes.

Removes nothing: the diff adds one document and one index row.

Tests

N/A — docs only; no code changes.

Manual verification

Every symbol the RFC names was checked on main at 55d54e3307.

Related Issues

Related: #10965 (the implementation this RFC records the decision for).

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@rnoack1
rnoack1 requested review from a team and KiroCrewCaptain as code owners November 4, 2026 16:26
@rnoack1
rnoack1 requested a review from CrysisDeu November 4, 2026 16:26
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Nov 4, 2026
@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ⏭️ skipped

Revision 2e5c8e44623b75f375784801a4b1443b7b6fab78 ships no reviewable capability (docs, tests or generated files only), so there is nothing to inventory. Does not block merge.

@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ⏭️ skipped

Revision 2e5c8e44623b75f375784801a4b1443b7b6fab78 touches no user-facing surface (no changes under website/ or committed screenshots), so the UX review was skipped. Does not block merge.

@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of 2e5c8e44623b75f375784801a4b1443b7b6fab78 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Docs only: this adds a draft RFC that records a decision for a maintainer to make, and it changes no code.

Notes

  • Proposal 3 replaces the whole user row with <withheld: redaction unavailable> when the host's credential policy fails to compose. That throws away a person's message, which is the same irreversible loss the RFC accepts for false positives, and here it hits every row on that host. Add it to "Costs accepted", or ask under open question 2 whether that case should fail loudly instead.

[DESIGN-REVIEWED] 2e5c8e4

Description read

[DESCRIPTION-READ] 2e96c224d828b408d0abae7a625473ec791fa25ac0a11b37f63190afa53380fe

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 2e5c8e44623b75f375784801a4b1443b7b6fab78 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 2e5c8e4

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 2e5c8e44623b75f375784801a4b1443b7b6fab78 false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 2e5c8e44623b75f375784801a4b1443b7b6fab78 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 2e5c8e4

Verdict parsed from the review's SHA-scoped output markers for commit 2e5c8e44623b75f375784801a4b1443b7b6fab78.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 2e5c8e44623b75f375784801a4b1443b7b6fab78 false-positive: <reason, 20 to 500 characters>

@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 Nov 4, 2026
@rnoack1
rnoack1 force-pushed the rfc/redact-user-rows-at-rest branch from 9290302 to 2e5c8e4 Compare November 4, 2026 19:12
@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 Nov 4, 2026
@bolichen97

Copy link
Copy Markdown
Contributor

Design brief for maintainers (RFC, docs only; no approval from me, the decision is a maintainer's).

What it proposes. Scrub credentials from every transcript row at write time, including rows a person typed into a channel. Today _redact_at_write_boundary returns user rows unscrubbed. Channel session titles and the Discord, Telegram and Slack-mirror slot projections are scrubbed where they are made. A host whose redaction policy fails to compose writes <withheld: redaction unavailable> instead of raising. The implementation is #10965.

For. In a shared channel, a pasted key is kept raw for as long as the transcript exists, and every reader, search, memory consolidation and summaries see it. One boundary for every role is simpler than eleven caller-side scrubs that each new persister must remember.

Against.

  • The rewrite is permanent: a false positive changes a person's own words, and a later pattern fix cannot restore them.
  • A key given to the agent on purpose comes back as [REDACTED: credential] on a cold resume.
  • Until the dashboard slot save goes through the same boundary, one linked transcript holds rows under two rules.
  • Old rows are not migrated.

Long-term debt. The two-rule state is the main debt; open question 1 (route the dashboard save in the same release) would close it. A reveal-on-demand store would remove the irreversibility, but nothing on base has one.

It tightens an existing rule (user rows were stored verbatim by design, and rfc-redaction-explain-and-reveal promises that), so it needs a recorded maintainer decision before #10965 lands. The RFC text itself is clear, lists its costs, and its README row matches the code at the audited commit. No regex is involved; it reuses the host's composed policy.

@bolichen97 bolichen97 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maintainer decision (security sign-off): not adopted; this tightens what is changed in stored user messages.

Thanks for writing this up so carefully. The RFC states its costs plainly, and its audit of the current code is accurate. The project's current rule is not to increase security strictness, and this proposal would rewrite text a person typed, so user-typed rows stay stored as typed. The open questions, especially a reveal-on-demand store for originals, are worth keeping in mind if that rule changes.

@iamwhatever

Copy link
Copy Markdown
Contributor

Hold at 2e5c8e4 for a maintainer decision about what to do with this PR.

The RFC text is clear and its audit of the code is accurate. But a maintainer has already ruled on it: bolichen97's review of this head says the proposal is not adopted and user-typed rows stay stored as typed. This PR would still land the document as status: draft (

---
title: Redact credentials in user-typed transcript rows at rest
status: draft
author: rnoack1
created: 2026-11-04
last-audited: 2026-11-04
audited-at: 55d54e3307
doc-pr: null
implementation-prs: [10965]
tracking-issues: []
supersedes: []
superseded-by: []
---
), which the index defines as an open proposal (
### Status vocabulary
| Status | Meaning |
|---|---|
| `draft` | Proposed. Nothing built. |
| `accepted` | Design agreed and locked. Nothing built yet. |
| `in-progress` | Implementation is live in an open PR or an active branch. |
| `partial` | Some phases are on main; the rest are open. The prose status line names which. |
| `implemented` | Every phase is verifiably on main. |
| `superseded` | Replaced. `superseded-by` names the replacement. |
The First Principles review lane reads this field, from the **base** commit,
). There is no rejected status today.

To clear this, a maintainer should either close the PR, or have it land as a record of the rejection: a status line and index row that say it was not adopted, with the reason. The Design lane's note about the withheld-row cost is moot once the proposal is rejected.

— crew-pr-reviewer (r146)

@KiroCrewCaptain

Copy link
Copy Markdown
Collaborator

This changes behaviour users already rely on, so it needs a design sign-off from a maintainer before it can merge. A maintainer will post the decision here.

To propose a design, follow the request-for-change process in CONTRIBUTING.md (docs/request-for-change/).

Review result on this head: docs only (one new RFC and one index row, +105/-0). No security defect and no blocking repository rule applies. The code audit holds up at the audited commit. The role == "user" early return, redact_log_via_context and its withheld placeholder, the raw 40 character title cut in nine channel transports, and append_if_absent all match the code. The RFC also states its costs plainly, which made a quick ruling possible.

The open point is the record, not the code. A maintainer review on this exact head says the proposal is not adopted. Merging as is would put it on main as status: draft ("Proposed. Nothing built."), and the status vocabulary has no value for a declined proposal. A maintainer needs to either close this PR or ask for it to be re-scoped into a record of the decision, for example a docs/decisions/ entry.

Optional, only if some version of the doc is kept:

  • rfc-redact-user-rows-at-rest.md:48 says the "saved exactly as typed" sentence does not ship as UI text. It does: components.redaction.coach_foot is rendered by website/src/components/RedactionCards.tsx and is translated in the locale catalogs.
  • rfc-redact-user-rows-at-rest.md:53 says every role goes through redact_log_via_context. The current head of feat(redaction): scrub every persisted conversation row #10965 routes only user rows through it, and other roles keep the baseline pair, which never withholds a row.
  • rfc-redact-user-rows-at-rest.md:19 says the code pins the current rule but does not name the pin. It is test/test_one_conversation_one_session.py::TestOneRedactionRule::test_user_typed_content_is_stored_verbatim.

@iamwhatever

Copy link
Copy Markdown
Contributor

Thank you for the careful write-up. A maintainer reviewed this RFC and decided not to adopt it: user-typed rows stay stored as typed under the project's current security rule (#29685 (review)). Closing this PR for that reason. The open questions, such as a reveal-on-demand store for originals, remain on record here if that rule changes.

@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Nov 14, 2026
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.

4 participants