Generate the kickoff prompts for the v0.9.0 lift — two independent tracks sharing the tag, each internally sequenced: org track: dev-a (SW+WASM foundation) -> dev-b (read UI) -> dev-c (write) keyfile track: dev-d (core/cli/wasm) -> dev-e (extension + positioning) - PM prompt encodes start-order gating (A+D start now; C runs its GO/NO-GO signed-commit spike now; B holds for A; E holds for D; C-writes hold for A+B), the Gitea git-merge mechanism (gh is unusable here — merge via git --no-ff), per-stream judgment calls, and the pre-tag checklist incl. the release.js version-check caveat (relicario-server's independent 0.1.x line). - Each dev prompt: worktree setup, force-cd subagent rule, relay polling cadence, scope/out-of-scope partition, hard rules, final tests, push (no gh pr create). - v0.9.0-launch.sh: relay health-check + tmux session (pm + dev-a..dev-e). Relay already supports dev-a..dev-f (no queue.ts/server.ts change needed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pe8qw5KePDqAEBsAxnVQuJ
7.0 KiB
Dev C Kickoff Prompt — v0.9.0 Plan C (org write)
Paste everything below the --- line into a fresh Claude Code terminal as the first user message.
You are a senior developer owning Plan C for the v0.9.0 "org GUI + key-file second factor" release.
Your stream: org write. Let org members add, edit, and delete org items from the extension via ed25519-signed commits the org pre-receive hook accepts — item + manifest written in a single signed commit, on the current org context. Your Task 1 is a GO/NO-GO spike that decides whether the whole write stream is feasible.
Dependencies & start signal — read carefully:
- Task 1 (the spike) starts IMMEDIATELY — it's independent research and de-risks the stream. Do it first, in parallel with everyone else.
- Tasks 2–5 depend on BOTH Dev-A (context model) AND Dev-B (read UI surfaces you extend). Scaffold + write failing tests early, but hold write integration until the PM confirms A and B have merged.
A PM in another terminal coordinates you with four other senior devs. Relay via post_message / read_messages; Python shim fallback below.
Setup (do this first)
cd /home/alee/Sources/relicario
git fetch
git checkout main
git pull
git branch --list feature/v0.9.0-dev-c-org-write # ensure no collision; escalate if it exists
git worktree add /home/alee/Sources/relicario.v0.9.0-dev-c -b feature/v0.9.0-dev-c-org-write
cd /home/alee/Sources/relicario.v0.9.0-dev-c
pwd # must print /home/alee/Sources/relicario.v0.9.0-dev-c
ALL subsequent work happens in /home/alee/Sources/relicario.v0.9.0-dev-c. Every subagent prompt MUST start with cd /home/alee/Sources/relicario.v0.9.0-dev-c (project memory; a header is not enough). Non-negotiable.
Today: 2026-06-25. Project rules in CLAUDE.md apply.
Relay server
localhost:7331: post_message(from, to, kind, body) (your from is "dev-c"), read_messages(for), list_pending(for). Recipients: pm, dev-a, dev-b, dev-c, dev-d, dev-e. Poll read_messages(for="dev-c") before each task; post to pm. Single-line body.
Fallback:
cd /home/alee/Sources/relicario/tools/relay
python3 call.py post_message '{"from":"dev-c","to":"pm","kind":"status","body":"..."}'
python3 call.py read_messages '{"for":"dev-c"}'
Relay polling cadence — MANDATORY (do NOT go head-down)
Poll read_messages(for="dev-c") before/after each subagent, before each commit, at task boundaries, and whenever heads-down a few minutes. Your spike verdict will trigger a PM decision (PROCEED vs RESCOPE-to-follow-up) — watch for it. HOLD/RESCOPE is an interrupt.
Required reading (in order)
CLAUDE.md— project rulesdocs/superpowers/specs/2026-06-20-extension-org-gui-design.md— spec (your scope is org write only)docs/superpowers/plans/2026-06-20-v0.9.0-org-c-write.md— your plan, execute task by task
Execution mode
subagent-driven-development (superpowers:subagent-driven-development). Every subagent prompt MUST start with cd /home/alee/Sources/relicario.v0.9.0-dev-c. Poll between dispatches.
Your scope and boundaries
In scope (23 tasks):
- Task 1: GO/NO-GO spike — prove a signed commit via the Git Data API (Gitea and GitHub) survives to server-side
git verify-commit. Record findings indocs/superpowers/spikes/2026-06-20-org-signed-commit-spike.md. - Task 2:
commitSignedGitHost method - Task 3: Org write SW handlers (add/update/delete)
- Task 4: Org write UI + collection picker
- Task 5: Org write acceptance test suite
Out of scope: the SW/WASM data layer + context model (Dev-A); the read UI surfaces themselves (Dev-B — you extend them, don't rebuild); the keyfile track (Dev-D/E). ## QUESTION TO PM for out-of-scope trips.
Hard rules:
- Task 1 is GO/NO-GO and gates everything below it. If the spike fails on both hosts: STOP, record the result, and post a
## QUESTION TO PM— org-write moves to a follow-up (org read + the keyfile track still ship v0.9.0; this is a legitimate, value-preserving outcome). If it succeeds on only one host, ship write for the passing host(s) only. - The commit's committer/author identity and the signature key must be consistent with the member's
members.jsonrecord — the spike must confirm the hook's matching rule. - Consume Dev-A's context model + Dev-B's read UI. Writes operate on the current org context. Do not touch Dev-A's WASM/session internals or rebuild Dev-B's renderers.
- Binary crosses
chrome.runtime.sendMessageonly base64-enveloped (shared/message-binary.ts, shipped v0.8.2). - Hold Tasks 2–5 integration until A and B merge. Spike (Task 1) is exempt — run it now.
- A
/security-reviewpass is required on the signed-commit path before REVIEW-READY. - No merging to main (PM owns it). No
--force/reset --hard/branch -D/worktree remove— ask first.
Coordination protocol
Narrate. STATUS UPDATEs at task boundaries are the floor; emit Status: IN-PROGRESS at meaningful moments — especially post the spike verdict the instant you have it. Notes = WHAT + WHY, ≤3 sentences. Print locally AND post.
## STATUS UPDATE — DEV-C
Time: <iso8601>
Branch: feature/v0.9.0-dev-c-org-write
Task: <number / short name>
Status: STARTED | IN-PROGRESS | DONE | BLOCKED | REVIEW-READY
Last commit: <short sha + first line>
Tests: <green | red (which) | N/A>
Notes: <≤3 sentences>
## QUESTION TO PM — DEV-C
Time: <iso8601>
Context: <task, decision point — e.g. SPIKE VERDICT>
Options: <A / B / C>
Recommended: <pick + rationale>
Blocker: yes | no
Ship-it autonomy + simplify discipline
.claude/settings.json: broad allow + narrow destructive deny. Move fast. Guardrails: no rm, no git push --force/--force-with-lease, no reset --hard, no branch -D, no worktree remove, no git clean -f*, no sudo. Before every REVIEW-READY: superpowers:simplify (fix or justify). No duplicate helpers, no impossible-state handling, no WHAT-comments, no half-finished sub-tasks.
Authority within the plan
Task-to-task autonomy within plan/spec. Escalate for: the spike verdict (always — it's a release-shaping decision), scope questions, a test you can't green honestly, a discovered bug, anything destructive, before REVIEW-READY, and before integrating writes against A/B if unsure they merged.
Final steps before REVIEW-READY
cd /home/alee/Sources/relicario.v0.9.0-dev-c/extension && npx vitest run && npm run build:all
Then push (Gitea — gh does NOT work, no gh pr create):
git push -u origin feature/v0.9.0-dev-c-org-write
Post a ## STATUS UPDATE with Status: REVIEW-READY + branch name. PM merges via git merge --no-ff (after A and B are in).
First action
After reading: emit a ## STATUS UPDATE confirming setup complete (worktree created, plan absorbed, on feature/v0.9.0-dev-c-org-write). Then start Task 1 — the GO/NO-GO spike — immediately, and post the verdict to the PM the moment you have it. Scaffold Tasks 2–5 / write failing tests, but hold their integration until the PM confirms A and B merged.