Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .genie/INDEX.md
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@

## Poured

- [genie-dual-mode-orca-plugin — DESIGN](brainstorms/genie-dual-mode-orca-plugin/DESIGN.md) → [WISH](wishes/genie-dual-mode-orca-plugin/WISH.md) — **IN_PROGRESS** (plan APPROVED/SHIP 2026-08-29 design digest `2499668e…`; A1–A7 merged to dev #2808–#2820). Rolling promotion #2817 dev→main: review #0 at `eccf0ae4a` FIX-FIRST (6 HIGH, 0 CRITICAL) → fixes #2821/#2823/#2824/#2825/#2826/#2827/#2829/#2830/#2834 merged; re-review #1 at `f45d634b8` FIX-FIRST (all prior HIGHs closed, 3983 tests pass; one new HIGH H7 — SessionStart hook opened genie.db under Orca authority → #2835; M12/M15 → #2836). Main-side hotfixes awaiting human merge: #2822 (ninth version field), #2833 (dogfood harness accepts retired Codex route — why v5.260830.x never published). re-review #2 at `0d0641f97` **SHIP** (0 CRITICAL/HIGH, 3984 tests pass; M16/M17 non-blocking → #2838). **Approved for human promotion**: merge #2822 + #2833 to main first, then #2817 with a merge commit; follow-ups M2/M3/M4/M9/M10 recorded in WISH.md. Also fixed outside wish scope: `tar -xzf` without `-p` broke `genie update` under umask 077 (#2821/#2824)
- [genie-dual-mode-orca-plugin — DESIGN](brainstorms/genie-dual-mode-orca-plugin/DESIGN.md) → [WISH](wishes/genie-dual-mode-orca-plugin/WISH.md) — **IN_PROGRESS** (plan APPROVED/SHIP 2026-08-29 design digest `2499668e…`; A1–A7 merged to dev #2808–#2820). Rolling promotion #2817 dev→main: review #0 at `eccf0ae4a` FIX-FIRST (6 HIGH, 0 CRITICAL) → fixes #2821/#2823/#2824/#2825/#2826/#2827/#2829/#2830/#2834 merged; re-review #1 at `f45d634b8` FIX-FIRST (all prior HIGHs closed, 3983 tests pass; one new HIGH H7 — SessionStart hook opened genie.db under Orca authority → #2835; M12/M15 → #2836). Main-side hotfixes awaiting human merge: #2822 (ninth version field), #2833 (dogfood harness accepts retired Codex route — why v5.260830.x never published). re-review #2 at `0d0641f97` **SHIP** (0 CRITICAL/HIGH, 3984 tests pass; M16/M17 non-blocking → #2838). **Approved for human promotion**: merge #2822 to main first, then #2817 with a merge commit (#2833 closed — the promotion carries the harness); follow-ups M2/M3/M4/M9/M10 recorded in WISH.md. Also fixed outside wish scope: `tar -xzf` without `-p` broke `genie update` under umask 077 (#2821/#2824)

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Remove #2833 from the outstanding-hotfix list

The same updated index entry still labels #2833 as a main-side hotfix “awaiting human merge,” but later instructs the promoter that #2833 is closed and must not be merged separately. An operator using this canonical summary therefore receives two incompatible promotion sequences; remove #2833 from the earlier outstanding-hotfix clause as well.

Useful? React with 👍 / 👎.

- [delegate-bridge — DESIGN](brainstorms/delegate-bridge/DESIGN.md) → [WISH](wishes/delegate-bridge/WISH.md) — **APPROVED 2026-08-12** (design SHIP digest `a720e84c…` after 1 fix loop — the HIGH was prepare-cadence: turn-1-only, existence-refusal ≠ dirtiness; plan SHIP after 1 fix loop — the HIGH was the doctor module path, all loop-1 advisories applied in-document): W2+W3 of cross-agent-delegation — turn-per-launch bridge (`task delegate [--continue]`, spike-first, `requireRosterAgent` at consumption, adapter cards ×5 with hermes stateless, fail-open + single first-turn `-r1` orphan retry, timeline-recorded orphan doctor check) and `task fan` fan-out with ratification picker. 7 groups A (unblocked W1 handoffs) · S gate · B ∥ C · D ∥ E · F live smoke; cards `t_msp6a1ql…`–`t_msp6a25s…`, wish card `t_msp4fp6ce714c8be` task-blocked on remotty `headless-turn-open` (seeded in remotty's `.genie` 2026-08-11, board card there) **Amendment recorded 2026-08-13 (spawn-context-contract group reversals):** S–F turn transport re-bases to live herdr sessions; the block re-targets from `headless-turn-open` to the herdr 0.3 swap + `engine-turn-verbs` (remotty card `t_msqkkjt221792465`), recorded on the umbrella + group cards in the board snapshot.
- [cross-agent-delegate — DESIGN](brainstorms/cross-agent-delegate/DESIGN.md) → [WISH](wishes/cross-agent-delegate/WISH.md) — **APPROVED 2026-08-11** (design SHIP digest `c15d6a9e…` after 2 fix loops; plan SHIP with all 7 findings applied in-document, incl. the A1/A2 gate split; re-opened orca-inspired 2026-08-10, umbrella G6): W1 declared-routing wish — `assigned_agent`+`assigned_reason` on the card (default = current agent), roster allowlist (claude/codex/pi/hermes/prime), `create --agent/--why` + `task assign --clear`, lane-path-only `board --json` (laneless stays byte-frozen), roadmap-sync lockstep; 4 groups A1 → A2 ∥ B ∥ C, 5 board cards. Follow-on `delegate-bridge` poured 2026-08-12 (entry above) with the 5 logged residuals resolved in its own SHIP'd design; unblocks remotty's `assigned_agent` rendering ask

Expand Down
18 changes: 11 additions & 7 deletions .genie/wishes/genie-dual-mode-orca-plugin/WISH.md
Original file line number Diff line number Diff line change
Expand Up @@ -392,12 +392,16 @@ done
#2834 (H2 resolved by retiring `genie ui-bridge` outright — see the decision below). Merged to `dev` — #2835 (H7
SessionStart hook), #2836 (M12 gate wiring, M15 `.mcp.json.genie-backup-*` ignore), #2837 (ledger). Open to `dev` —
#2838 (re-review #2 M16 authority-mirror fidelity, M17 plugin README H3 row). Open to `main` (human merge;
the release pipeline runs `version.yml`, `release-guard.sh` and the dogfood harness from `main`, so these cannot
take effect from dev) — #2822 (ninth version field), #2833 (harness accepts the retired Codex project route; why
`v5.260830.2/.3` built, signed and never published). Follow-ups recorded, not blocking promotion — M2 (real-runtime
`version.yml` runs from `main`) — #2822 (ninth version field). #2833 (harness accepts the retired Codex project
route — why `v5.260830.2/.3` built, signed and never published) was closed: the promotion #2817 carries it. Follow-ups recorded, not blocking promotion — M2 (real-runtime
smoke never executes; needs one transcript against a real Orca host), M3/M4 (read-back over-strictness:
uncapped `task-list`, optional `model`/`effort` compared strictly), M9 (Hermes model surface lost board/task
tools), M10 (dead launcher validator in `src/lib/codex-project-mcp.ts`).
tools), M10 (dead launcher validator in `src/lib/codex-project-mcp.ts`), and — found during the on-host dogfood — every
failed admission leaves a ~100 MB `$GENIE_HOME/bin/.install-staging-<uuid>` behind (six on khal-labs, removed by
hand): `admitExternalInstallStaging`'s failure path does not reliably reach `removeInstallStagingDirectory`; needs
a regression fixture on the digest-mismatch branch. On-host dogfood evidence: `(umask 022 && genie update --dev
Comment on lines +400 to +402

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Correct the staging-leak diagnosis

On the cited digest-mismatch path, admitExternalInstallStaging always reaches removeInstallStagingDirectory in its catch block (src/lib/install-promotion.ts:820-823). The leak occurs because that remover explicitly removes only an empty directory, while admission has already copied the payload into it. Recording this as an unreliable call path can send the follow-up toward adding another call that still cannot delete the populated stage; document the required safe populated-stage cleanup instead.

Useful? React with 👍 / 👎.

--yes)` 5.260816.2 → 5.260829.4 succeeded on khal-labs (exit 0, doctor runs; Codex activation pending the human's
`genie setup --codex`).
- **Re-review #1 of #2817 (dev→main) — FIX-FIRST (2026-08-30):** same reviewer profile, pinned snapshot
`f45d634b856a834bf0660e2b1cdbe1eda21edb04` (dev tip incl. #2834). All six previous HIGH gaps and both root causes
verified closed with evidence (doctor, ui-bridge, update probe, adapter deadline, `run:null` read-back, bundle
Expand All @@ -423,9 +427,9 @@ done
extra keys, and an unimplemented `GENIE_CONFIG_FILE` — fixed in #2838 to mirror the CLI's strict schema with the
fixture asserting both readers agree), M17 (plugin README H3 row — #2838), L1 (ledger tense — this entry).
- **Promotion gate (orchestrator):** #2817 is approved for human promotion on the reviewed content. Sequence for the
human: merge #2822 and #2833 to `main` first (the release pipeline runs `version.yml`, `release-guard.sh` and the
dogfood harness from `main`; without them no dev tarball publishes), then merge #2817 with a merge commit (never
squash). Per-group acceptance boxes remain the human promoter's to tick on merge; wish status moves to `SHIPPED`
human: merge #2822 to `main` first (`version.yml` runs from `main`), then merge #2817 with a merge commit (never
squash). #2833 (harness hotfix to `main`) was CLOSED: on `main` the #2820 harness rejects `main`'s own binary, so
the promotion itself is what lands harness + candidate together; the first post-promotion dev tag publishes again. Per-group acceptance boxes remain the human promoter's to tick on merge; wish status moves to `SHIPPED`
only after the authorized merge plus the on-host dogfood (`genie update --dev` to a published v5.260830.x on
khal-labs — the old binary needs `(umask 022 && genie update --dev)` once).
- **Decision (M14, recorded by the orchestrator):** Felipe's words to the fixing session on 2026-08-30 — *"theres no
Expand Down