ワンクリックで
continue-prp
Continue executing a phased PRP from the next phase, restoring context from the latest checkpoint.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Continue executing a phased PRP from the next phase, restoring context from the latest checkpoint.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Safely upgrade this Cortex repo to the latest template: detects on-disk state, reconciles infrastructure, and applies idempotent migrations — preserving your customizations and never overwriting your edits without asking.
Safely upgrade this Cortex workspace to the latest template: detects on-disk state, reconciles infrastructure, and applies idempotent migrations — preserving your customizations and never overwriting your edits without asking.
Maintainer auto-check for the Cortex source repo. Invoke this PROACTIVELY and automatically WHENEVER files under templates/ have been added, edited, deleted, renamed, or restructured — and before committing template changes — to classify the change as ADDITIVE (handled by the runner's reconciliation) or MIGRATION-NEEDED (breaking/structural), and scaffold a migrations/NNN-*.md if one is required. Do not wait to be asked. For use INSIDE the template source repo only; it is never shipped to user projects.
Run a full project health check across context-engineering, the knowledge base, and docs to surface gaps, drift, and inconsistencies.
Validate implementation progress of a PRP against its original requirements, auditing completed, current, and remaining phases.
Validate a generated PRP structure against its source requirements to catch gaps and misalignments before execution begins.
| name | continue-prp |
| description | Continue executing a phased PRP from the next phase, restoring context from the latest checkpoint. |
| argument-hint | <prp-path> |
| disable-model-invocation | true |
Continue working on a phased PRP. This command is for resuming work on a complex feature after closing a previous context.
This command assumes the PRP folder already exists (created by /generate-prp) — UNLESS
$ARGUMENTS is a single .md file (a simple PRP), handled by STEP 0 below.
This command handles three resume scenarios, routed by $ARGUMENTS in STEP 0:
/checkpoint;.md file) — resumed from its checkpoint.The phased flow (STEP 1 onward) serves BOTH (1) and (2) and is unchanged; /checkpoint writes the
resume state this command reads back.
Look at $ARGUMENTS:
context-engineering/PRPs/FEATURE-NAME) → Phased PRP → follow the
"PHASED PRP" flow below (STEP 1 onward).context-engineering/PRPs/feature-name.md) → Simple PRP → follow the
"SIMPLE PRP RESUME" section immediately below, and SKIP the phased steps.context-engineering/_STATUS.md for workspace context.$ARGUMENTS) completely — especially the status header at the top and the
## 🔁 CHECKPOINT — RESUME STATE section at the bottom (written by /checkpoint). If there is
no checkpoint section, fall back to the Test Execution Tracker + Success Criteria to infer
remaining work.NEXT ACTION. Honor its Decisions (do not re-litigate) and
Environment / gotchas so you don't re-derive or undo settled work.TodoWrite to track
progress. Log any new fixes and decisions back into the file as you go./checkpoint $ARGUMENTS to refresh the checkpoint. When the PRP is truly
finished, set the status header to Complete, finish the Test Execution Tracker, and (if tracked)
update its entry in context-engineering/_STATUS.md.Do not look for $ARGUMENTS/_STATUS.md, phases, or HANDOFF.md for a simple PRP — those exist
only for phased (folder) PRPs.
Read context-engineering/_STATUS.md to understand:
Read {$ARGUMENTS}/_STATUS.md to find:
Read files in this order:
{$ARGUMENTS}/OVERVIEW.md - Full feature contextHANDOFF.md (if not Phase 0) - What was done beforePLAN.md - Tasks to executeTEST-CASES.md - Test cases for guided testing after implementationFIXES.md - Check for pending fixesCOMPLETED.md - Check what's already done (if In Progress)Based on the status in _STATUS.md:
PLAN.mdCOMPLETED.md as you workCOMPLETED.md to see what's donePLAN.mdCOMPLETED.md as you workTEST-CASES.md for this phaseFIXES.md for pending issuesFIXES.md with solutions (root cause, code changes)_STATUS.md to next phase and continueFor each task in PLAN.md:
COMPLETED.md:
Use TodoWrite tool to track your progress through tasks.
After completing tasks:
PLAN.mdAfter validation commands pass, run the guided testing walkthrough:
TEST-CASES.md========================================
READY FOR GUIDED TESTING
========================================
I'll walk you through {N} test cases for Phase {X}.
For each test, I'll:
- Tell you what to do
- Validate the result after your action
- Mark pass/fail
Let's start with TC-A1: {title}
========================================
When phase tasks are complete:
Update _STATUS.md:
**Status:** Awaiting Testing
Update the phase emoji in Quick Status:
- Phase {N}: Awaiting Testing
Tell user:
========================================
PHASE {N} TASKS COMPLETE
========================================
Status: Awaiting Testing
Please test the functionality and let me know:
- If it works → say "approved" and I'll complete the handoff
- If issues → describe the problem and I'll fix it
========================================
When user reports issues:
Add to FIXES.md:
## Fix #{N}: {Issue Title}
**Reported:** {Date}
**Status:** Pending
**User Feedback:**
> {Exact quote from user}
**Investigation:**
{What you found when investigating}
Implement the fix
Update FIXES.md with:
**Root Cause:**
{Why this happened}
**Solution:**
{What was changed}
**Files Changed:**
| File | Line(s) | Change |
|------|---------|--------|
| `src/xxx.ts` | 42-45 | {description} |
**Code Changes:**
// BEFORE {old code}
// AFTER {new code}
**Status:** Fixed
When user approves (says "approved", "looks good", "works", etc.):
Fill in HANDOFF.md with:
REQUIRED for EVERY phase (not just final phase):
Knowledge can be lost between sessions. Update documentation after EVERY phase completion to preserve context.
Run the full /update-knowledge-base process, which will:
Update Knowledge Base:
_SUMMARY.md files and knowledge-base/INDEX.mdUpdate Project Documentation:
active-projects/{project}/AGENTS.md only if this work changed a durable, every-session fact: project structure, tech stack, a build/test/lint command, a new coding convention, an "always do X / never do Y" rule, or a high-level architecture/data-flow change. Litmus test: "Would a fresh agent need this in EVERY session, regardless of the task?" If no, do not touch AGENTS.md.AGENTS.md. Those belong in the PRP's own FIXES.md/COMPLETED.md and the knowledge base (gotchas/, decisions/, flows/, implementations/{project}/).Why every phase matters:
Skip documentation updates only if:
For detailed instructions, see /update-knowledge-base.
Update {$ARGUMENTS}/_STATUS.md:
**Current Phase:** Phase {N+1} - {Next Phase Name}
**Status:** Not Started
## Quick Status
- Phase {N}: Complete
- Phase {N+1}: Not Started
If this was the FINAL phase of the feature:
context-engineering/_STATUS.mdTell user:
========================================
PHASE {N} COMPLETE
========================================
Completed: {Phase Name}
Handoff: {path to HANDOFF.md}
Next: Phase {N+1} - {Next Phase Name}
You can close this context now.
To continue later: /continue-prp {$ARGUMENTS}
========================================
┌─────────────────────────────────────────┐
│ Read _STATUS.md → Find current phase │
│ Read OVERVIEW.md → Feature context │
│ Read previous HANDOFF.md → What's done │
│ Read current PLAN.md → What to do │
│ Read current TEST-CASES.md → Tests │
│ Read FIXES.md → Pending issues │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ Execute tasks from PLAN.md │
│ Update COMPLETED.md as you go │
│ Run validation commands │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ Guided Testing Walkthrough │
│ Read TEST-CASES.md │
│ Walk user through each test case │
│ Validate after each action │
│ Mark pass/fail in tracker │
└─────────────────────────────────────────┘
↓
┌────────┴────────┐
↓ ↓
┌─────────┐ ┌─────────────────┐
│ ALL PASS│ │ FAILURES FOUND │
│ │ │ → Log in FIXES │
│ │ │ → Fix issues │
│ │ │ → Retest failed │
└────┬────┘ └────────┬────────┘
│ │
│ (loop until all pass)
↓ │
┌─────────────────────────────────────────┐
│ Write HANDOFF.md │
│ Update _STATUS.md → Next phase │
│ Context can be closed │
└─────────────────────────────────────────┘
The Goal: Seamless continuation. You should be able to pick up exactly where the previous session left off, with all knowledge preserved in the knowledge base and project documentation.
AGENTS.md for project-specific safety rulesgit commit, git push, or similar commands