Repository navigation
docs: RISK.md declares which changes are high risk - #9194
Conversation
Quest-Claim: f34b4ba427abd058
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Approach: sound. It adds a RISK.md in quest's format and makes RELEASE.md#releases the one definition of an API break and a wire change.
Reviewed changes
Docs only. I confirmed that RELEASE.md defines both terms, so the new links resolve to what the text says.
RISK.md: sets the tier prose and a six-path## Highlist.quest/README.md: replaces the two restated definitions of breaking with links toRELEASE.md#releases.CONTRIBUTING.mdandREVIEW.md: point atRISK.md.quest/m1/risk-policy.md: cut down to the work that remains, now waiting only on the guardrails quest.
claude-sonnet-5-5 | 𝕏
|
No actionable comments were generated in the recent review. WalkthroughThe changes add Merge Risk:
|
|
Maintainer pick (2026-11-19, /quest-complete): ✅ /quest-merge at ab15a7c (rejected: read it first; iterate on the tiers). (Written by Claude Opus 5.5 with high) |

Implements part of quest/m1/risk-policy.md and rewrites it to what remains.
Problem
quest reviewgives every PR a risk tier since moq-dev/quest#716, but a repository without aRISK.mdrecords every change as medium. That makes moq's tiers useless, and the maintainer's week of reading tiers (moq-dev/quest's risk-tier quest) cannot cover this repository.Approach
RISK.mdinquest's format (pinned d4df7d5): prose the classifier reads, plus a## Highlist of paths whose every change is high.-wipthe next cut publishes, hang catalog JSON, container formats, CLI flags, config keys, env vars); what a token grants; when an agent may merge or post without the maintainer; release packaging and publish scripts.RELEASE.md,version.json,.release-plz.toml,sh/version.py,sh/api-diff/,sh/gh/release.sh.questalready holdsRISK.md,.github/workflows/, and.github/actions/.sh/dispatch.sh,sh/changed.sh), other agent instructions. Low: tests, benchmarks, TLA+ models, quest plans, examples, demos, docs.RISK.mdcites it too.CONTRIBUTING.md's landing gate andREVIEW.mdpoint atRISK.md.just gate) and the classes that land without a pick.quest's format has no place for either yet; moq-dev/quest's guardrails quest adds both, so the quest now waits on that alone. Sized XS.Impact
No public API or wire change.
quest reviewtiers on PRs reviewed after this merges change from a flatmediumto the classifier's tier. Nothing acts on a tier yet.Alternatives
rs/*/src/**,rs/hang/src/catalog/): rejected. Most of the tree is public API, so a path rule would make nearly every PR high. Whether a change breaks is judged from its diff, which is the classifier's job.packaging/by path: rejected for prose. They are spread acrosssh/<lang>/with tests beside them, so a glob would also catch test edits.questand add nothing local, and the guardrails quest defines that format.Decisions
Taken without the maintainer (AFK); each is easy to undo in review:
rs/moq-auth,js/auth, the relay's token checks) is high when it changes what a token grants. The Goal did not list it. A widening there is a vulnerability in every relay after the next cut.RISK.md. Other agent-instruction edits are medium.sh/dispatch.sh, 47 merges last month) is medium, not held high: it changes too often, and a miss reachesmain, not users.RELEASE.mdis held by path: agents follow it to cut and backport (0 merges last month).RELEASE.md, which already defines both. This is the "one source the others cite" the quest's Plan asked for.Validation
quest check(with and without--base origin/main) passes. Appending a malformed## Highitem (/bad) made it fail withRISK.md:56: ... starts with /, confirming it parses this file and accepts all 6 items.quest review --base origin/quest/m1/risk-policy --harness codex --effort low, Codexgpt-6.1-sollow, about 15.9k input tokens), on throwaway local branches based on this branch, deleted afterward:--drain-timeout/MOQ_DRAIN_TIMEOUT: high, "CLI flags and environment variables are explicitly high risk".mainsince 2026-10-19, 163 touchedquest's built-in paths and 166 touched those plus this file's## Highlist (git pathspec count), so the list adds about 0.1%. The classifier's share comes on top.quest gate just checkpasses.just gate, no Rust cache) took 10m51s on 2026-11-18, against the scoped queue median of 5.5 minutes. Earlier runs' job data returned HTTP 502.Follow-ups
None new. The rest of the work stays in quest/m1/risk-policy.md, waiting on moq-dev/quest's guardrails quest.
Models
🤖 Generated with Claude Code
(Written by Claude Opus 5.5 with high)
Review
Rewritten by
quest reviewafter each round.Head: ab15a7c
Base: origin/main at 8e1605c
Risk: high,
it changes RISK.md, which is always high: it sets the tier of every later changeClassified: high by the always-high paths
Reviewer: Codex, gpt-6.1-sol, medium effort
Session:
01a1f4bb-fd27-73f1-99fc-803c27b44938, round 1Score: 9/10,
The risk policy matches existing release definitions, remaining guardrail work is explicitly deferred, and no actionable defects or blockers were identified; builds and tests were not run as instructed.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none