T-PDDK-gs-scaffold-package-and-tooling

Status: closed/done · Impact: high · Complexity: small

Goal

Stand up packages/ts/graph-scheduler/ as a standalone bun + moon workspace package with the same tooling shape as packages/ts/intersect/, so every later task has a place to add code and tests. No scheduling logic yet.

Today

Greenfield: packages/ts/graph-scheduler/ does not exist yet. The relevant context is the workspace root and the reference package this scaffold mirrors.

Location	Role today
packages/ts/	Bun workspace directory; root package.json globs packages/ts/*. Holds dashboard-client and intersect today — no graph-scheduler/.
packages/ts/intersect/	Reference TypeScript package whose tooling shape (package.json, tsconfig.json, moon.yml, .gitignore, README.md, index.ts) this scaffold mirrors.

Proposed

A compiling, testable, zero-runtime-dependency package skeleton: package.json (name graph-scheduler, type: module, exports map for ., ./graph, ./tracker, ./pull), tsconfig.json, moon.yml, biome/lint wiring matching the repo, a bun test entry that runs (zero tests is fine), an index.ts re-export stub, a README.md skeleton, and .gitignore.

Approach

1. Study packages/ts/intersect/ (package.json, tsconfig.json, .gitignore) and copy its tooling shape.
2. Create packages/ts/graph-scheduler/ with package.json, tsconfig.json, moon.yml, .gitignore, empty index.ts, README.md.
3. Wire it into the bun workspace (root already globs packages/ts/*); confirm bun install resolves it.
4. Confirm bun test packages/ts/graph-scheduler exits clean (no tests yet).

Files to touch

Location	Kind	Change
packages/ts/graph-scheduler/package.json	new	Package manifest + exports map
packages/ts/graph-scheduler/tsconfig.json	new	TS config mirroring intersect
packages/ts/graph-scheduler/moon.yml	new	moon project wiring
packages/ts/graph-scheduler/index.ts	new	Root re-export stub
packages/ts/graph-scheduler/README.md	new	README skeleton
packages/ts/graph-scheduler/.gitignore	new	Ignore build artifacts

Acceptance criteria

• packages/ts/graph-scheduler/ exists with package.json, tsconfig.json, moon.yml, index.ts, README.md.
• exports declares ., ./graph, ./tracker, ./pull.
• bun install resolves the workspace package; bun test packages/ts/graph-scheduler runs and exits 0.
• No runtime dependencies in package.json.

Out of scope

• Any scheduling logic, tests, or fixtures (later tasks).

Dependencies

• none

Post-mortem

Captured by /sdlc:task-work on 2026-07-04. PR: pending.

Acceptance criteria coverage

• AC-1 (packages/ts/graph-scheduler/ exists with package.json, tsconfig.json, moon.yml, index.ts, README.md): auto — files present, bunx tsc --noEmit clean.
• AC-2 (exports declares ., ./graph, ./tracker, ./pull): auto — verified in packages/ts/graph-scheduler/package.json; each subpath points at a real stub module.
• AC-3 (bun install resolves the workspace package; bun test packages/ts/graph-scheduler runs and exits 0): agent-manual — ran bun install (workspace symlink + bun.lock entry created) and bun test packages/ts/graph-scheduler (exit 0, 1 pass).
• AC-4 (no runtime dependencies): auto — no dependencies/devDependencies field in package.json.

What worked

• The packages/ts/intersect/ reference package (post-M-0010 migration) was an exact tooling template — package.json/tsconfig.json/moon.yml/.gitignore copied cleanly to a coherent zero-dependency skeleton.
• start_task.ts’s reset onto the fresh origin/main tip surfaced that the reference had migrated (plugin/lib/intersect → packages/ts/intersect) before the implementer used the stale path — the worktree reflected reality, not the divergent local checkout.
• bunx tsc --noEmit and the package smoke test gave fast, deterministic AC verification independent of the noisy repo-wide gate.

Friction and automation gaps

• Quality baseline captured against the working tree, not origin/main’s tree — sdlc quality baseline capture <sha> runs verbs against the caller’s cwd while labelling the output with origin/main’s SHA. Here local main was 2 commits ahead of and corpus-divergent from origin/main (a completed intersect migration), so Step 7’s baseline-gated diff produced 12 false new-drift: lines (migration-related closed-task OK lines, count summaries, a non-deterministic bun-test PID/port line). Fix: capture the baseline against the resolved origin/main tree (ephemeral worktree at the SHA) so the label matches the tree. → T-7FPF-baseline-capture-against-origin-main-tree
• bun test exits 1 on zero test files — the spec said “a bun test entry that runs (zero tests is fine)” but AC-3 requires exit 0, which bun does not give with no test files. Required adding a trivial smoke test. Fix: scaffolding specs that assert bun test … exits 0 should require ≥1 smoke test, or the AC should tolerate bun’s no-tests exit.
• autonomy: autonomous/pr gap-fill is discarded by start_task’s reset — the auto-defined ## Today section lands on the task branch, but start_task.ts’s reset --hard origin/main wipes it, so the readiness stamp reached origin/main while the required Today section did not (this PR re-adds it branch-side to reconcile on merge). Fix: the ensure-ready autonomy gate’s auto-define body edits should land on origin/main (like the stamp does), not on the soon-to-be-reset task branch. → T-RNHP-autonomy-body-fill-lands-on-main
• No plugin verb lands arbitrary task-body edits on origin/main — task-work Step 5 says to land body edits on origin/main before the gate, but only frontmatter has a --commit-on main path (ensure_ready_mutate/start_task). Landing the Today body edit needed an ad-hoc git push origin HEAD:main, which the sandbox auto-mode classifier denied (while the plugin’s own subprocess --commit-on main pushes were allowed). Fix: a --commit-on main body-edit primitive, or route the autonomy Today-fill through the existing main-commit path. → primitive in flight as T-NXAM-task-doc-edit-commit-on-main; autonomy wiring folded into T-RNHP-autonomy-body-fill-lands-on-main
• Pre-existing bun test failures on origin/main add gate noise — 2 stale sdlc --help goldens (missing the get-spawn-policy config op that another change added without regenerating top-help.txt) and 1 task_auto_define test that needs an origin remote its temp repo lacks. Not introduced here, but they muddy the quality signal.

Spawned follow-up tasks

• T-7FPF-baseline-capture-against-origin-main-tree (https://github.com/sksizer/dev/pull/663) [open/ready] — capture the quality baseline against the resolved origin/main tree, not the caller’s working tree. Spawned. Distinct axis from the in-flight baseline-dir/superproject PR cluster (there the failure mode is baseline not found; here it is false new-drift: when local main has diverged).
• T-RNHP-autonomy-body-fill-lands-on-main (https://github.com/sksizer/dev/pull/665) [open/ready] — route the ensure-ready autonomy auto-define body edits to origin/main so start_task’s reset preserves them. Spawned; merges the “auto-define body discarded by reset” bullet with the “--commit-on main body-edit primitive” bullet, building atop the in-flight primitive T-NXAM-task-doc-edit-commit-on-main.
• The bun test exits-1-on-zero-tests bullet and the pre-existing-bun test-failures bullet were SKIPPED (non-actionable / pure observation of pre-existing failures).

---

← Back to Tasks