Skip to content

docs(rfc): propose a default approval mode for new dashboard chats - #36369

Open
RohanK6 wants to merge 1 commit into
mainfrom
crew/conductor/rfc-default-approval-mode
Open

RohanK6 wants to merge 1 commit into
mainfrom
crew/conductor/rfc-default-approval-mode

Conversation

@RohanK6

@RohanK6 RohanK6 commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Record a proposal, for a maintainer to accept or change, that lets the owner choose the approval mode a new dashboard chat starts on.

Every plain dashboard chat starts on Normal. A person who works at Reads or Trust re-picks the mode in the composer for each new chat, and again after a restart. #6812 has 8 reactions and three duplicates; the one implementation attempt (#8984, closed) stalled on whether Trust may be a stored default, and the maintainer asked for a request-for-change on 2026-11-05.

Why it matters

Without a recorded decision, every implementation of #6812 re-opens the same question (which modes, which chats, which store), and nothing ships. An accepted RFC lets phases 2-3 land as small, reviewable PRs.

Not a goal

  • No code, config key or Settings row: this PR is docs only.
  • Trust as a stored default for plain chats: left to the maintainer (RFC section 6), phase 4 only if accepted.
  • Channels, cron, subagents, apps, crewmate chats and agent-created sessions: they keep how they start today.
  • Restoring a chat’s mode after a restart (Per-chat trust dies on gateway restart; armed loops and workers stall #12275).

What changed (motivation → approach → change)

  • Adds docs/request-for-change/rfc-default-approval-mode.md (status draft) and its row in docs/request-for-change/README.md.
  • The proposal: one owner setting, Normal or Reads, stored in config.json (a sealed, read-only leaf inside the sandbox, so the agent cannot pick the owner’s next start mode). It is applied only to a plain chat the owner opens from the dashboard, through seed_slot_approval with caller chat_default_mode, the seeder the crewmate Profile and Slack code-channel start modes already use. The seeder clamps to the approval_modes ceiling, never ends a declared YOLO grant and never replaces a posture the slot already holds; apply_approval_mode is rejected because a Reads or Trust press through it ends a YOLO grant.
  • Section 2 lists what exists today, every reference read at 0feaa8be90. Section 4 splits the work into phases that each ship alone. Section 7 lists the three questions for the maintainer.

Backwards compatibility

Compatible: docs only; no code, key or behaviour changes.

Removes nothing: the diff adds one RFC file and one index row.

Tests

Proof: not provable -- docs or config only

scripts/docs-lint.sh passes.

Manual verification

N/A — docs only; every code reference in the RFC was re-read at 0feaa8be90.

Real-machine verification

N/A -- the diff adds one RFC document and one RFC index row; nothing in it runs.

Related Issues

Part of #6812

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

Contribution License Agreement

Contributions are accepted under the project’s Apache-2.0 license.

@RohanK6
RohanK6 requested review from a team and KiroCrewCaptain as code owners November 13, 2026 03:48
@RohanK6
RohanK6 requested a review from mbriones98 November 13, 2026 03:48
@github-actions github-actions Bot added the readiness: action required A blocking check or review needs attention label Nov 13, 2026
@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed d59f4bd7527e10366f28e58a9a9d067d47fe3c69 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] d59f4bd

Verdict parsed from the review's SHA-scoped output markers for commit d59f4bd7527e10366f28e58a9a9d067d47fe3c69.

False positive or not applicable? A repository writer can comment:
/ai-review override fable d59f4bd7527e10366f28e58a9a9d067d47fe3c69 false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ⏭️ skipped

Revision d59f4bd7527e10366f28e58a9a9d067d47fe3c69 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 13, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of d59f4bd7527e10366f28e58a9a9d067d47fe3c69 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] d59f4bd

False positive or not applicable? A repository writer can comment:
/ai-review override gpt d59f4bd7527e10366f28e58a9a9d067d47fe3c69 false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of d59f4bd7527e10366f28e58a9a9d067d47fe3c69 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

This is a draft RFC and changes no code, so nothing here blocks it. It reuses the existing seed_slot_approval seeder and keeps the setting in config.json, which sandbox_leaves.py already makes read-only to the agent.

[DESIGN-REVIEWED] d59f4bd

Description read

[DESCRIPTION-READ] 3e91e6159baa26dde58f9d79ec376930b56fd5fa7a81700e5e79fb283234529c

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 13, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ⏭️ skipped

Revision d59f4bd7527e10366f28e58a9a9d067d47fe3c69 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 added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Nov 13, 2026

@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.

Direction signed by maintainer (bolichen97) for phases 1-3 only: a Normal / Reads start mode for the owner's plain dashboard chats. Phase 4 (Trust) is not signed here.

I read the full diff at 2cde0fd. My answers to section 7:

  1. Yes. A normal / trust_reads default in config.json, applied through seed_slot_approval at the owner's plain POST /api/chat/slots, is acceptable as phases 2-3. It reuses the seeder the crewmate and Slack code-channel defaults already use. The ceiling clamps it, it leaves a YOLO grant and an existing posture alone, and it records its own SEL caller. The store sits in _CREW_READONLY_LEAVES, so an agent cannot set its owner's next start mode. The default stays normal, so nothing changes until the owner opts in. A bad, missing or unreadable value reads as normal, and a failed seed never fails the create. Every other creation path in the 3.2 table keeps how it starts today.
  2. Trust (phase 4) is left open for a separate decision. Please don't add it to the key in phases 2-3.
  3. Once. The proposed one-time composer line when the ceiling clamps a Reads default is right.

For phase 2, please keep apply_approval_mode out of the create path exactly as 3.3 says, and pin with a test that a declared YOLO grant survives opening a new chat with the default set to Reads.

Not checked: "Dependency License Gate" is red on this head. This is a docs-only change, and I did not read that log.

@KiroCrewCaptain

KiroCrewCaptain commented Nov 13, 2026 •

Copy link
Copy Markdown
Collaborator

Code change approver recommendation: A person should decide

Why: the decider raised 2 possible problems and estimates a 55% chance that this change introduces a defect; a person should check them.

I'm requesting changes on this docs-only RFC: the boundary it draws around who gets the stored start mode doesn't match the gate it names. The default is meant to apply only to the machine owner's plain dashboard chats, and the RFC's table says a Shared Crew teammate's chat stays on Normal. But the gate it names, is_owner_dashboard_request, also lets teammates through, including on POST /api/chat/slots. The same is true of the "owner-gated" config API, so a teammate could also write the new key. A phase 2 built from §3.2 as written would start teammates' chats on Reads, where read-only commands run without an approval card. The fix is small: name is_host_owner_dashboard_request, the check the config-patch handler already uses for admission keys, and say who may write the key. The rest of the RFC checks out against the code (sealed config.json, the seeder clamping to the ceiling and never ending a YOLO grant, the existing crewmate and Slack code-channel seeds). CI and the AI review lanes all passed, but none of them caught this.

Things to check (possible problems the decider raised; at this confidence many turn out to be fine):

  • The RFC states the trust boundary for who gets the auto-approving start mode, then names a gate that does not enforce it. A phase-2 PR that follows §3.2 as written would start every Shared Crew teammate's new chat on the owner's Reads default, so read-only shell commands and first-party read MCP too (docs/request-for-change/rfc-default-approval-mode.md, medium)
  • The store argument covers the agent but not a Shared Crew teammate, who could set the machine owner's start mode for every future chat through the same config API. That may be acceptable, since teammates can already edit a crewmate's Profile approval_mode, but the RFC should say so or put the key be (docs/request-for-change/rfc-default-approval-mode.md, low)

Tier: 2, held for a person: design-doc-gate: D.

An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit d59f4bd752 · 191 changed lines. If this recommendation is wrong, reply here: replies are how the approver is evaluated.

@KiroCrewCaptain

Copy link
Copy Markdown
Collaborator

The design for phases 1 to 3 is signed off by a maintainer in the review on this head, so this PR now waits for a maintainer approval.

Review result: nothing blocks. The diff is docs only. It adds the new RFC and one row in the request-for-change index, and changes no code. The review checked the RFC's code references against the commit it names and against current main, and they hold. These include the slot's Normal start state, the clamp in seed_slot_approval, its rule never to end a YOLO grant or replace a posture the slot already has, the YOLO side effect of apply_approval_mode, and the read-only sandbox status of config.json. Done well: the 3.2 table names every creation path that keeps its current start mode, and that sets clear limits for phase 2.

Optional:

  • Record the acceptance in the RFC. Line 17, the status: draft front matter and the index row still say acceptance is requested. They could say accepted for phases 1 to 3 with phase 4 open, and include the section 7 answers.
  • Section 3.3 could say that the phase-2 seed, like the crewmate seed, waits on approval_mode_policy.verdict_pending() before it clamps. It should report default_mode_refused only on a final verdict, so a ceiling that is still installing does not show a wrong refusal.
  • The phase-2 row in section 4 could list its test pins. Examples: YOLO survives a create with a Reads default, a ceiling that denies Reads leaves Normal and still returns 2xx, a crewmate chat gets its Profile seed, and a failed seed still creates the slot.

@RohanK6
RohanK6 enabled auto-merge (squash) November 13, 2026 18:40
@RohanK6

RohanK6 commented Nov 16, 2026

Copy link
Copy Markdown
Contributor Author

Hello maintainers: this RFC PR is ready for review. At 2cde0fd readiness passes and auto-merge is armed; the only red lane is the Dependency License Gate, which this docs-only change does not touch. It proposes the Trust step for #6812, and its open questions are listed in the RFC itself. Thank you for taking a look.

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Nov 19, 2026
@RohanK6
RohanK6 force-pushed the crew/conductor/rfc-default-approval-mode branch from 2cde0fd to 4ce43d1 Compare November 19, 2026 17:17
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: passed Eligible automated validation passed for the current revision labels Nov 19, 2026
@RohanK6
RohanK6 force-pushed the crew/conductor/rfc-default-approval-mode branch from 4ce43d1 to 699eb20 Compare November 19, 2026 18:18
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Nov 19, 2026
Every plain dashboard chat starts on Normal, and people who work at
Reads or Trust re-pick it each time (#6812, 8 reactions, three
duplicates). The maintainer asked for a request-for-change. Propose
an owner setting, Normal or Reads, stored in the sealed config.json
and seeded through seed_slot_approval, the seeder the crewmate and
Slack code-channel start modes already use, only for a plain chat the
owner opens from the dashboard. The approval_modes ceiling clamps it
and a YOLO grant is left alone. Crewmate chats, channels, cron,
subagents, apps and agent-created sessions keep how they start; Trust
is left to a maintainer decision.

Part of #6812
@RohanK6
RohanK6 force-pushed the crew/conductor/rfc-default-approval-mode branch from 699eb20 to d59f4bd Compare November 19, 2026 21:19
@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 19, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

readiness: passed Eligible automated validation passed for the current revision

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants