菜单
Efficiently read and analyze pi agent session JSONL files. Use when asked to "read a session", "analyze a session", "what happened in this session", or given a .jsonl session file path.
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
Write messages, comments, PR reviews, Slack posts, and other prose in Sean's voice. Use when drafting anything that will be sent under Sean's name — code review comments, postmortem feedback, Slack messages, GitHub comments, coaching notes, design feedback, strategic takes. Triggers: 'write like me', 'in my voice', 'draft a comment for me', 'sound like me', 'sharpen this', 'this still sounds like AI'.
Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/PRD asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
Implement a piece of work based on a spec or set of tickets.
Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets, and resolve them one at a time until the way to the destination is clear.
Move todos through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs. Maps to pi's built-in todo tool; see ../todo-tracker.md for the full tracker → tool mapping.
Turn the current conversation into a spec and publish it as a todo — no interview, just synthesis of what you've already discussed.
| name | session-reader |
| description | Efficiently read and analyze pi agent session JSONL files. Use when asked to "read a session", "analyze a session", "what happened in this session", or given a .jsonl session file path. |
Parse pi session JSONL files into readable, structured output. Sessions live in ~/.pi/agent/sessions/<project>/ as .jsonl files.
Resolve the session file path. Sessions are stored at:
~/.pi/agent/sessions/--<path-with-dashes>--/<timestamp>_<uuid>.jsonl
If the user provides a partial path or project name, find the file:
ls -t ~/.pi/agent/sessions/*<project>*/*.jsonl | head -5
Always start with the overview to understand the session before diving deeper. Script paths below (scripts/read_session.py, references/session-format.md) are relative to this skill's directory — resolve them against the folder containing this SKILL.md when running the command.
uv run scripts/read_session.py <path> --mode overview
This shows: session metadata (model, project, cost), turn count, and a summary of every turn with timestamps and tool calls used.
Based on what's needed, use the appropriate mode:
| Goal | Command |
|---|---|
| See user/assistant conversation only | --mode conversation |
| See everything including tool I/O | --mode full |
| See what tools were called and results | --mode tools |
| Analyze token usage and costs | --mode costs |
| See subagent delegations with task/status/cost/paths | --mode subagents |
For large sessions, use --offset and --limit to page through user turns:
# Skip first 3 user turns, show next 5
uv run scripts/read_session.py <path> --mode conversation --offset 3 --limit 5
Control content truncation with --max-content:
# Show full tool outputs (no truncation)
uv run scripts/read_session.py <path> --mode full --max-content 0
# Shorter previews (500 chars per block)
uv run scripts/read_session.py <path> --mode full --max-content 500
When a session contains subagent calls, the --mode subagents output shows paths to each subagent's own JSONL session. Read those with the same script:
# Persistent artifact copy (always available)
uv run scripts/read_session.py ~/.pi/agent/sessions/<project>/subagent-artifacts/<hash>_worker.jsonl --mode overview
# Temp session file (may be cleaned up)
uv run scripts/read_session.py $TMPDIR/pi-subagent-session-<id>/run-0/<timestamp>.jsonl --mode overview
Subagent sessions use the exact same JSONL format. The overview and full modes all handle subagent data — they show inline summaries with agent, model, cost, duration, and status for each subagent run.
When summarizing a session for the user, include:
If you need to understand the raw JSONL format (for custom parsing), read:
references/session-format.md
The critical thing to know: message content is nested at line.message.content, NOT line.content. Content is always an array of typed objects (text, toolCall, thinking). Tool results are separate message entries with role: "toolResult".