[namastex888]

Implements the launch-removal execution group of spawn-context-contract (Wave 5 — the replacement surface genie context --plan already exists on dev, and the wish-scoped spawn side is held as a remotty PR per plan).

What ships

• genie launch deleted atomically with its consumers — src/term-commands/launch.ts + src/lib/v5/warp-launch.ts and their tests removed; resolveWorktreesBase relocated to src/lib/v5/launch-worktrees.ts (mechanical import-only change in doctor-worktrees.ts, whose retarget stays owned by the drain group). CLI registry is now 16 commands (was 14 documented / 17 actual).
• $work multi-session mode retired with the recorded amendment (DRAFT U2 → Option A): native subagents are the only dispatch mode; the spawn plan preview is genie context --wish <slug> --plan; supervised parallel sessions are arranged from the spawn side.
• Hermes genie_work_plan rewired to genie context --wish <slug> --plan [--group g] (single group; schema groups[] → group); pi-genie mirror updated in lockstep. Hermes contract tests green (58 passed).
• .warp/.mcp.json scaffolding removed from genie init — registerProjectMcpConfigs now writes only .mcp.json + the marker-owned .codex/config.toml; the file-pinned paths in launch.ts + codex-project-mcp.ts are gone; symlink-guard and preflight tests updated. The .genie/launch/ gitignore rule stays as legacy residue protection.
• Zero stale references — git grep -l 'genie launch' skills/ plugins/ == 0; sync-plugin-skills --check green; skills + all three plugin mirrors updated (work, lifecycle, native-surfaces, genie-hacks, pm, review, council, rules, dispatch-contract, codex-integration-map).
• Honest command counts — README/CLAUDE.md now say 16 commands with context, idea, and ui-bridge rows; AGENTS.md freeze rule no longer names the deleted verb; scripts/release-docs.test.ts derives the command inventory from the live --help registry instead of a hardcoded list (carried order).
• e2e lifecycle — the launch --dry-run section is replaced by a context --wish <slug> --plan section proving: one line of versioned JSON, composed branch + full 40-hex base SHA + all three task ids, exit 0, byte-identical genie.db and unchanged .genie/ file set modulo the documented -wal/-shm VFS pair (read-only proven), no worktrees materialized.

Not touched (out of scope by design)

• Doctor launch-residue check/--fix (genie.ts doctor description + doctor-worktrees.ts + doctor-modes.ts docstrings) — retargeted last by the drain group.
• src/lib/interactivity.ts historical comment; .genie/ tracked brainstorm/wish records are historical documents.
• Hook manifests; hook policy logic (only the git-freeze-guard deny-reason UX text naming the deleted verb changed, plus its test — deny behavior untouched).

Validation evidence

• bun run check (clean env, umask 022): all gates green; full suite 3323 pass / 4 fail, where the 4 failures are the recorded known-red baseline classes, not new: doctor 120ms latency (~1972ms measured), updateCommand publication (network-bound release flake), 2× exact GitHub release asset reconciliation timeouts (network-bound release flake; 13/14 pass on isolated re-run).
• Baseline recorded on origin/dev under identical env before any change: 6 failures (same doctor + updateCommand classes, plus the Codex-absent codex-app-server integration cluster — now passing on this branch) and the noisy-env run (36 failures, FORCE_COLOR stderr cluster + same classes). No new failure class; zero failures in any file this PR touches.
• bash tests/e2e/v5-lifecycle.sh — PASS locally (macOS) and on the ubuntu CI leg (sidecar-tolerant file-set bound, mirroring context.test.ts).
• Hermes: uv run --with pytest --with pyyaml --no-project python -m pytest plugins/hermes-genie/tests -q — 58 passed.
• Sweep: test "$(git grep -l 'genie launch' skills/ plugins/ | wc -l | tr -d ' ')" -eq 0 — passes.
• bun scripts/sync-plugin-skills.ts --check — OK (22 physical skills, byte-identical mirror).

Named follow-ups (recorded, out of scope)

• GENIE_WORKTREES_DIR + doctor residue semantics retire with the drain group's retarget (this PR only relocates the helper).
• Nothing else open.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 0370b0de-6545-43c5-b0c8-cfcc92d951f2

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7bea11bbb3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Do not present every ready task as the current wave

For wishes with dependency-ordered groups, this command does not preview the current wave: context returns every task whose database status is ready, while this same skill notes at line 107 that CLI-created tasks are ready from birth and that waves must instead be derived from the WISH.md Execution Strategy. A spawn consumer treating the advertised output as the wave plan can therefore dispatch later groups before their prerequisites; describe this as spawn-context data and direct callers to derive the wave first and request each selected --group.

Useful? React with 👍 / 👎.