feat(#79, #80): add task-ID convention, partial delivery, and mode routing#82
Conversation
Task IDs (T1, T2, ...) are scoped to issues and appear in commit messages as #<id>/<Tn>. Partial-delivery PRs use "Addresses #N" instead of "Closes #N" to ship completed tasks while keeping the issue open for remaining work. Adds: task ID format to naming convention, partial-delivery PR template, partial delivery rules to closure-and-review, commit format with task refs, retrieval by task ID. Made-with: Cursor
Each phase now has a recommended execution mode (plan or agent) alongside its tier. Plan mode prevents premature execution during brainstorming, discovery classification, PR synthesis, knowledge retrieval, and reviews. Claude Code agents can self-switch via exit_plan_mode. For Codex and Cursor, mode routing is advisory using the same confirm/warn/off config. Adds: phase-to-mode mapping, runtime capability table per tool, mode advisory in phase entry check, mode routing prompt format, README sections for mode routing and task-level delivery. Made-with: Cursor
|
Reviewed-by: openai/gpt-5.4 (codex, effort: high) Finding:
|
…ty-neutral advisory Addresses PR #82 review finding: `/plan` and `Shift+Tab` were stated as Codex user controls but openai/codex#11180 only establishes that agent-initiated switching is missing — it does not document those controls. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
[shiplog/amendment] #80: address review finding — Codex mode-switch controlsFinding: PR review flagged that Fix applied in
Cursor Authored-by: claude/opus-4.6 (claude-code) |
|
Reviewed-by: openai/gpt-5.4 (codex, effort: high) No findings. The task-ID, partial-delivery, and mode-routing changes stay consistent across the skill, reference docs, templates, and README. Local merge simulation against |
Summary
Add mode routing to shiplog phases and introduce the task-ID convention for incremental delivery. Analysis phases (brainstorming, discovery classification, PR synthesis, knowledge retrieval, reviews) now recommend plan mode; execution phases recommend agent mode. Claude Code agents can self-switch to plan mode; Codex and Cursor get advisory prompts using the existing routing config.
Also introduces task IDs and partial-delivery PRs from #79 as the mechanism enabling this incremental shipment.
Addresses #79 (completes minimal task-ID convention: T1 task ID format, T2 commit format)
Addresses #80 (completes T1 phase-to-mode mapping, T2 runtime capability model, T3 phase entry check wiring, T5 README update; T4 partial — Claude Code self-switch only)
Completed Tasks
From #79 (task-to-git mapping — minimal subset)
T1,T2, ...)feat(#42/T1): ...)Addresses #Ninstead ofCloses #N)git log --grep="#42/T1")From #80 (mode routing)
model-routing.mdSKILL.mdRemaining Tasks
#79 (design work deferred)
taskfield#80
EnterPlanMode/ExitPlanModeself-switch instructions — blocked on upstreamBlocked On
EnterPlanMode/ExitPlanModetools for agent-initiated mode switching in Codex CLIJourney Timeline
Initial Plan
Mode routing was identified as a gap: shiplog routes by model tier but says nothing about execution mode (plan vs. agent), despite all three target tools supporting the distinction. Issue #80 was created. The partial-delivery pattern from #79 was needed to ship this incrementally since T4 (Codex self-switch) depends on an upstream feature.
What We Discovered
exit_plan_modetool — agents CAN self-switch to plan modeKey Decisions Made
confirm/warn/offwith tier routingexit_plan_modemay be internal/unstableT1, not#80-T1)Addresses #Ninstead ofCloses #NTesting
status: in-progressenvelope (notresolved)Knowledge for Future Reference
EnterPlanMode(openai/codex#11180), updatereferences/model-routing.mdcapability table and add Codex self-switch instructions to SKILL.md.exit_plan_modetool name comes from community documentation (ClaudeLog), not official Anthropic docs. If the API changes, update the capability table.taskfield).Authored-by: claude/claude-4.6-opus-high-thinking (cursor)
Captain's log — partial delivery by shiplog