fix(test): include test/e2e in test globs and build before tests

[robotlearning123]

Fixes #80.

WHAT

• package.json test script: add test/e2e/*.test.ts to both test globs so the 9 orphaned e2e tests in test/e2e/cli.e2e.test.ts actually run (57 -> 66 tests).
• Both test and test:coverage now run npm run build first: cli.e2e.test.ts spawns dist/index.js, which did not exist under CI-only test invocation (unbuilt: 2 pass / 7 fail; built: 9 pass / 0 fail).

WHY

• Issue test/e2e/ never runs: package.json glob misses the subdirectory (9 orphaned tests) #80: package.json:32 glob 'test/*.test.ts' expands to only 4 top-level files; test/e2e/ never executes. Reproduced at origin/main e704b49.
• Two explicit globs (test/.test.ts test/e2e/.test.ts, plain sh expansion) instead of a quoted 'test/**/*.test.ts': the recursive form over-collects 2 fixture sample files (test/fixtures/l3-repo/src/index.test.ts, test/fixtures/standard-repo/src/index.test.ts -> 68 tests, 2 of which fail without their fixture setup).

TEST EVIDENCE (all run in a fresh worktree at e704b49, npm ci exit 0)

• RED premise: sh -c 'echo test/.test.ts' -> 4 files, no test/e2e/; npx tsx --test test/.test.ts -> # tests 57 / # pass 57 / # fail 0
• GREEN: rm -rf dist && npm test -> exit 0, # tests 66 / # pass 66 / # fail 0
• Issue acceptance oracle: npm test 2>&1 | grep -qE '^# (pass|tests) 6[6-9]' -> ORACLE-PASS (exit 0)
• npm run test:coverage -> # tests 66 / # pass 66 / # fail 0
• npx tsx --test test/e2e/cli.e2e.test.ts -> # tests 9 / # pass 9 / # fail 0 (orphans were not stale; no skips, no deletions)
• git diff --stat: package.json | 4 ++-- only; zero overlap with PR fix(logger): route pino records to stderr so check --json stdout parses #89 files (src/utils/logger.ts, test/check-json.test.ts)

SCOPE

• Exactly 2 script lines in package.json. No CI, dependency, or doc changes. No e2e test deleted or skipped.

[robotlearning123]

STATUS: needs_work — independent review pending (grok-402 protocol).

Builder: devin (GOAL_COMPLETE, commit b7830b3). Verified by execution in fresh worktree at e704b49, npm ci exit 0:

• RED premise: sh -c 'echo test/.test.ts' -> 4 files, no test/e2e/; npx tsx --test test/.test.ts -> # tests 57 / # pass 57 / # fail 0
• GREEN: rm -rf dist && npm test -> exit 0, # tests 66 / # pass 66 / # fail 0
• Issue acceptance oracle: npm test 2>&1 | grep -qE '^# (pass|tests) 6[6-9]' -> ORACLE-PASS (exit 0)
• npm run test:coverage -> # tests 66 / # pass 66 / # fail 0
• npx tsx --test test/e2e/cli.e2e.test.ts -> # tests 9 / # pass 9 / # fail 0 (orphaned tests are not stale; none deleted or skipped)
• git diff --stat origin/main..HEAD -> package.json | 4 ++-- (only file; zero overlap with PR fix(logger): route pino records to stderr so check --json stdout parses #89 files src/utils/logger.ts, test/check-json.test.ts)
• Commit message Conventional-Commits, no AI-attribution text (grep clean).

Review lane status (fleet-quota-gate, observed 2026-09-28T07:40:27Z): grok blocked until 2026-09-28T11:23:05Z; all review-role alternates also blocked (agy 2026-10-01T15:25Z, gpt6pro 17:00:09Z, opencode 17:23:20Z). Per the grok-402 protocol this PR stays draft until an independent cross-family review runs. NOT merge_ready.

One-command resume (run after 2026-09-28T11:23:05Z once fleet-quota-gate grok returns ok:true):

cd /tmp/loop-wt-agent-next_agent-ready-issue80 && grok -p 'Independent review of PR #90, head origin/loop/20260928_032029-issue80 (b7830b3), base main (e704b49). Diff is package.json only: test and test:coverage scripts gain npm run build && prefix plus test/e2e/*.test.ts glob. Run npm test (expect # tests 66 / # pass 66 / # fail 0) and the issue oracle npm test 2>&1 | grep -qE "^# (pass|tests) 6[6-9]". Check scope creep, attribution footers, and overlap with PR #89. Post verdict as a PR comment via gh pr comment 90 -R agent-next/agent-ready. Review-only: never merge, never approve, never push.' --always-approve --output-format plain

[robotlearning123]

Independent review gate status: REVIEWER-LANE-UNAVAILABLE (grok 402), needs_work on the gate.

grok lane blocked: 2 live probes 2026-09-28 both failed with API error (status 402 Payment Required): Grok Build usage balance exhausted. Per the no-silent-skip rule the independent review is NOT complete; verdict stays REQUEST_CHANGES until grok runs.

Execution verification done in /tmp/loop-wt-agent-next_agent-ready-issue80 @ b7830b3 (node v22.22.0, npm 10.9.4):

• diff origin/main...loop/20260928_032029-issue80: package.json only, +2/-2 (adds test/e2e/*.test.ts + npm run build && to test and test:coverage)
• baseline without e2e: npx tsx --test test/*.test.ts -> # pass 57
• full new suite: npm test -> # tests 66, # pass 66, # fail 0, # skipped 0 (all 9 previously-orphaned e2e tests pass once included; none stale)
• issue test/e2e/ never runs: package.json glob misses the subdirectory (9 orphaned tests) #80 acceptance oracle: npm test 2>&1 | grep -qE '^# (pass|tests) 6[6-9]' -> PASS (# tests 66, # pass 66)

Non-blocking note: test:watch (package.json:34) still uses the old single-dir glob without e2e; watch mode is dev-only and out of this issue's minimal scope.

One-command resume runbook (after grok top-up):
timeout 540 grok --always-approve --cwd /tmp/loop-wt-agent-next_agent-ready-issue80 -p "REVIEW the diff origin/main...loop/20260928_032029-issue80 on agent-next/agent-ready for correctness, test coverage, and scope creep. Read the actual diff (git diff). REPRODUCE blocking findings with commands. Output exactly: VERDICT: SHIP or VERDICT: FIX-FIRST, then findings[] with file:line + failure scenario + severity, then checked_does_not_hold[]. Default to FIX-FIRST when uncertain."

If the worktree was already cleaned up, recreate first: git -C /home/robot/workspace/agent-next/agent-ready worktree add /tmp/loop-wt-agent-next_agent-ready-issue80 loop/20260928_032029-issue80

[robotlearning123]

Verification (cycle 20260928_032029, issue #80): premise reproduced on main — package.json:32 glob expands to only 4 top-level files; npm test 57/57/0 while test/e2e/cli.e2e.test.ts holds 9 it() blocks that never ran. Fix at head b7830b3 (package.json, 2 script lines): npm run build && + two explicit globs (test/.test.ts test/e2e/.test.ts). Two honest refinements over the issue's literal proposal: (1) the e2e file spawns dist/index.js and CI never builds — without build it is 2 pass/7 fail, so the test script builds first; the 9 orphans are NOT stale (9/9/0 after build — none deleted or skipped); (2) the issue's quoted 'test/**/*.test.ts' over-collects 2 fixture sample files (68 tests) — explicit globs give exactly 66, satisfying the issue's acceptance oracle. Session re-verified by execution in a fresh worktree from clean dist: npm test 66/66/0, oracle grep ORACLE-PASS, coverage run 66/66 (84.65% lines, build rounds). Zero overlap with sibling PR #89 files. Independent review did NOT run (reviewer lane quota-blocked, retryAt 2026-09-29T05:10:50Z) — status: needs_work, review-gate only; zero code findings.

One-command resume when the reviewer lane recovers:
git -C /home/robot/workspace/agent-next/agent-ready fetch origin --prune && git -C /home/robot/workspace/agent-next/agent-ready worktree add /tmp/loop-wt-agent-next_agent-ready-issue80 loop/20260928_032029-issue80 2>/dev/null; timeout 540 grok --always-approve --cwd /tmp/loop-wt-agent-next_agent-ready-issue80 -p "REVIEW origin/main...loop/20260928_032029-issue80 on agent-next/agent-ready for correctness, test coverage, and scope creep. Read the actual diff (git diff). Judge against issue #80. Output exactly: VERDICT: SHIP or VERDICT: FIX-FIRST, then findings[] each with file:line + concrete failure scenario + severity, then checked_does_not_hold[]."

[robotlearning123]

Reviewer lane recovered — grok review result: VERDICT SHIP.

findings[]: none. checked_does_not_hold (execution-verified):

• package.json:32 build-before-tests is required, not scope creep: running test/e2e/cli.e2e.test.ts without dist fails 7/9; with build, npm test passes 66/66 and the issue test/e2e/ never runs: package.json glob misses the subdirectory (9 orphaned tests) #80 oracle grep passes.
• Quoted 'test/**/.test.ts' glob (issue suggestion) would also match test/fixtures//src/index.test.ts and fail 2 tests; current unquoted expansion is correct.
• CI (ci.yml:74) and release.yml:38 both call npm test, so e2e is included in CI/release.
• test:coverage (package.json:33) shares the same globs; passes 66/66.
• test:watch unchanged is out of scope: test/e2e/ never runs: package.json glob misses the subdirectory (9 orphaned tests) #80's acceptance targets npm test.