fix(coding-agent): drop an exports entry with no module behind it

[10kH]

Found by checking every literal target in every package's exports / main / module / types / bin against the tree: 158 paths across the workspace, one broken pair.

Problem

"./extensibility/custom-commands/bundled/review": {
  "types":  "./src/extensibility/custom-commands/bundled/review/index.ts",
  "import": "./src/extensibility/custom-commands/bundled/review/index.ts"
}

That file does not exist — bundled/ contains only ci-green. The module was deleted by 56c8970c “refactor(coding-agent): trim bundled agents to four canonical role agents (#922)” and the exports entry was left behind.

Nothing in the tree imports the subpath, which is exactly why it survived: no import breaks to signal the drift. What it does affect is the published surface — the package declares a subpath that fails to resolve for consumers, and export-surface linters flag entries pointing at missing files.

The sibling is the control: ci-green has the identical shape and its file exists, so the pattern is right and only this entry is stale.

Guard

sdk-package-exports.test.ts already asserts specific subpaths are present or deliberately null. It now also walks exports, main, module, types and bin, skips null entries and wildcard patterns, and asserts every literal target resolves on disk.

Proof-first: restoring the stale entry fails that test, naming both offending mappings:

exports/./extensibility/custom-commands/bundled/review/types  -> ./src/.../review/index.ts
exports/./extensibility/custom-commands/bundled/review/import -> ./src/.../review/index.ts

Verification

bun run check in packages/coding-agent (whole-package biome + tsc): exit 0. sdk-package-exports.test.ts: 10 pass / 0 fail.

[Yeachan-Heo]

Signed exact-head verdict — PR #3306 @ 83c77a1d95e45b5732de2f0ab5834861ed12afbc

Verdict: APPROVE

Scope: Removes the stale ./extensibility/custom-commands/bundled/review exports entry from packages/coding-agent/package.json (the target file was deleted in 56c8970c / #922 while the manifest entry was left behind), and adds a generic guard test that walks exports/main/module/types/bin, skips null and wildcard (*) targets, and asserts every literal target resolves on disk.

Verification performed:

• Confirmed packages/coding-agent/src/extensibility/custom-commands/bundled/review/index.ts does not exist anywhere in history after 56c8970c; bundled/ contains only ci-green.
• Grepped the tree for other references to the removed subpath — none outside the now-fixed package.json.
• Confirmed the pre-fix manifest (current dev HEAD 1122e2ced) still contains the dangling ./extensibility/custom-commands/bundled/review entry, so the bug is real and reproducible on dev.
• Built native deps and ran bun test packages/coding-agent/test/sdk-package-exports.test.ts at the PR head: 10 pass, 0 fail, including the new "resolves every declared export target to a file that exists" test.
• New test correctly skips null values and *-containing wildcard patterns before calling fs.existsSync, matching the stated behavior.
• CHANGELOG entry accurately describes the fix.

No functional risk: this only shrinks the declared public export surface to match what already worked at runtime (nothing imported the dead subpath). Dev-branch CI at the PR's exact base (1122e2ced) is terminal/green, and the PR's own CI is green.

[Yeachan-Heo]

Verdict: APPROVE — exact head ccc17cc25c503b10090c71e191efba38ab7dede0

Rebase-only refresh of the prior exact-head review at 83c77a1d95e45b5732de2f0ab5834861ed12afbc (APPROVE).

Checks performed:

• Content identity confirmed, not assumed: diff <(git show 83c77a1d --format= -- packages/coding-agent/package.json packages/coding-agent/test/sdk-package-exports.test.ts) <(git show ccc17cc2 --format= -- packages/coding-agent/package.json packages/coding-agent/test/sdk-package-exports.test.ts) is empty — the PR's own commit is byte-for-byte identical to the version the prior review validated. CHANGELOG.md is intentionally excluded from this file set: its diff context shifted because independently-merged dev commits inserted new entries above the same insertion point, which is expected rebase churn, not a content change to this PR's own line.
• Rebase mechanics verified: git merge-base confirms this PR carries the same single commit both before and after the rebase; every additional commit between the two heads is an independently-merged dev commit (own-PR-numbered), not new work folded into this PR.
• Exact-head terminal CI: 15 success / 0 failure / 0 pending / 5 skipped (total 20). No pending/queued checks remain.

Prior technical analysis still applies verbatim to this head.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]