fix(gitignore): ignore .artifacts/ release-lane PR-CI evidence scratch dir (#1150, #1155)

[SisyphusZheng]

Problem

The v0.44.0-beta.1 publish run (autoflow-release.yml, run 33734627360) failed after all release-tier gates passed with:

Refusing release from a dirty worktree: ?? .artifacts/

The workflow downloads the authorizing PR-CI evidence artifact into .artifacts/pr-ci/ (workflow lines ~76–91), and assertCleanWorktree (tools/lib/git-cleanliness.ts) rejected the untracked .artifacts/ directory. The dry-run path cannot catch this because it returns before the cleanliness assertion. Publish blocker for #1150 / #1155.

Owner

v0.44 Kimi implementer, dispatch packet: release-lane fix — .artifacts/ scratch dir breaks publish-existing clean-worktree assertion (#1150 publish blocker).

Before

With .artifacts/pr-ci/pr-full-ci-evidence.json present (exactly what the release workflow downloads), git status --porcelain prints ?? .artifacts/, and filterNonEvidenceDirty returns ["?? .artifacts/"] — assertCleanWorktree refuses the release.

RED proof (base cbd60136):

$ git status --porcelain
?? .artifacts/
$ deno eval 'import { filterNonEvidenceDirty } from "./tools/lib/git-cleanliness.ts"; ...'
nonEvidenceDirty: ["?? .artifacts/"]
WOULD REFUSE: Refusing release from a dirty worktree   (exit 1)

After

.artifacts/ is listed in .gitignore under the existing "Root-local media/agent scratch artifacts (never release inputs)" section, with a one-line comment tying it to the release workflow's PR-CI evidence download dir. With the evidence file present, git status --porcelain is empty and deno task repo:hygiene passes.

GREEN proof (head 13d914c):

$ test -f .artifacts/pr-ci/pr-full-ci-evidence.json && echo present
present
$ git status --porcelain
(empty)
$ deno task repo:hygiene
Repo hygiene check passed.   (exit 0)

Why-not-second-owner

This change introduces no duplicate-looking implementation: it adds one ignore entry to the existing root-local scratch section of .gitignore. No new tooling, test framework, or allowlist logic was added (repo:hygiene has no gitignore-coverage test hook, so per the packet the behavior is proven directly instead of inventing one).

Evidence

All run against 13d914c on branch kimi/v044-release-artifacts-gitignore:

Command	Exit
deno task fmt:check (Checked 1552 files)	0
deno task lint (Checked 883 files)	0
deno task repo:hygiene (with .artifacts/pr-ci/ present)	0
deno task workflow:check	0
git status --porcelain with .artifacts/pr-ci/pr-full-ci-evidence.json present	0, empty output
deno eval filterNonEvidenceDirty pre-fix	1 (RED, would refuse)

Additional scan (packet step 3) of the release lane found no other unignored scratch writes: the workflow otherwise writes only $HOME/.npmrc (outside the repo); publish-existing persists only under docs/release/ (allowlisted in git-cleanliness.ts); npm pack tarballs packages/*/*.tgz are covered by *.tgz; pack-time temp dirs use Deno.makeTempDir (outside the repo); www/app/data/_generated- writes are allowlisted.

Scope

• .gitignore — 2 lines added (entry + comment) in the root-local scratch artifacts section.

Nothing outside the packet-owned path was edited. No product code, no workflow edits, no tooling logic changes.

Risk

None known. Residual: the implementer agent doc's generic "do not commit/push/PR" guardrail was overridden by this packet's explicit workflow step 4 (branch/commit/push/PR); noted for the record. Follow-up: re-run the v0.44.0-beta.1 publish after merge.

[cloudflare-workers-and-pages]

Deploying openelement with    Cloudflare Pages

Latest commit:	13d914c
Status:	✅  Deploy successful!
Preview URL:	https://8c1c8ddf.lessjs.pages.dev
Branch Preview URL:	https://kimi-v044-release-artifacts.lessjs.pages.dev

View logs

[github-actions]

APIError: Insufficient Balance

opencode session  |  github run