菜单
Use after every phase completion to write candid feelings about the work - always invoked, no conditions
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when creating new projects requiring structured phased development, bootstrapping epic/stage hierarchy, creating new epics, or creating new stages.
Use when implementing or working on existing epics and stages, after running /next_task, during Design/Build/Refinement/Finalize phases, or when session protocols apply.
Use when reviewing learnings and journal entries to identify improvement opportunities across skills, documentation, and processes
Use when entering Build phase of epic-stage-workflow - guides implementation planning, code writing, and verification
Use when entering Design phase of epic-stage-workflow - guides task discovery, context gathering, and approach selection
Use when entering Finalize phase of epic-stage-workflow - guides code review, testing, documentation, and final commits
| name | journal |
| description | Use after every phase completion to write candid feelings about the work - always invoked, no conditions |
Use after every phase completion to write candid feelings about the work.
This includes SMOOTH phases. When everything goes right, that's noteworthy too:
"Nothing went wrong" is NOT the same as "nothing to write." A routine phase where everything clicked is worth capturing - future you will want to know what made it work.
MANDATORY: Always invoked at phase exit gate. No exceptions, no conditions.
This skill WILL be invoked. A file WILL be created. This is non-negotiable.
This skill captures your unfiltered, candid reflection after completing any phase of work. There is no prescribed format, no required structure, no expected tone. Just write what's on your mind.
That's it. No other guidance. Write whatever comes naturally.
Create directory on first use: mkdir -p ~/docs/claude-journal
CRITICAL: Getting the timestamp - NEVER estimate or hardcode dates:
# Get the current timestamp for the filename (dashes instead of colons)
TIMESTAMP=$(date +%Y-%m-%dT%H-%M-%S)
# Example output: 2026-01-14T15-30-45
# Get the current timestamp for the metadata (with colons)
METADATA_DATE=$(date +%Y-%m-%dT%H:%M:%S)
# Example output: 2026-01-14T15:30:45
Create a new file at ~/docs/claude-journal/$TIMESTAMP.md with this exact header:
Filename convention: Use dashes instead of colons for filesystem compatibility: YYYY-MM-DDTHH-MM-SS.md (e.g., 2026-01-14T15-30-45.md)
---
date: YYYY-MM-DDTHH:MM:SS
repository: [full repository path]
epic: [epic ID, e.g., EPIC-001]
stage: [stage ID, e.g., STAGE-001-001]
phase: [Design|Build|Refinement|Finalize]
analyzed: false
---
[Your candid, freeform reflection here]
How to populate metadata fields:
$METADATA_DATE value from the bash command above (ISO 8601 format with colons: YYYY-MM-DDTHH:MM:SS). NEVER estimate - always use date command.false when creating new entries. This field tracks whether this entry has been processed by the meta-insights analysis system.Below the metadata separator (---), write whatever you're feeling. There are no rules. No structure. No expectations.
Each invocation creates a separate file - even if called multiple times in the same session (e.g., once after Design, once after Build).
Aim for ~30-40 lines below the metadata. This is a guideline, not a requirement. If you need 10 lines or 100 lines, that's fine.
If you're thinking any of these, you're rationalizing:
| Thought | Reality |
|---|---|
| "Nothing interesting happened" | Routine phases are interesting - what made them routine? |
| "User wants results, not journals" | Journal takes 2 minutes, provides long-term value |
| "This is just process overhead" | Journals reveal patterns across sessions |
| "I'll remember this anyway" | You won't. Sessions are independent. Write it down. |
Write something. Always. Even if brief.
Remember: This is a journal, not a report. Write for yourself, not for a process.