菜单
Find, inspect, summarize, or search prior pi sessions/conversations. Use when the user asks to refer to previous sessions, inspect session history for a working directory, search across all conversations, recover prior tool output, or audit what was stored.
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
Run a Multi-Character Code Council with four GPT-5.6 Sol reviewer characters and an xhigh skeptical chair. Use only when the user explicitly requests an MCC review, names the Multi-Character Code Council, or directly invokes this skill. Never use for a generic review, second opinion, adversarial review, cross-review, pre-implementation review, or request for multiple perspectives unless the user explicitly asks for MCC.
Capture durable resume context into a project-local agent state file. Use when the user asks to capture the current session, prepare a handoff before starting a new session, preserve resume context, or update saved agent state.
Read project-local saved agent state and orient before work. Use when the user asks to preflight, resume from a handoff, load saved agent context, compare saved state with the repository, or understand where prior work left off.
Generate new images and edit, restyle, or combine existing images with OpenAI GPT Image 2 through Codex's native subscription-backed image tool. Use when the user asks to create an image, illustration, poster, logo, icon, mockup, diagram, product shot, visual asset, or to modify one or more image files.
Use the local Gertrude `gt` task helper to list, spawn, review, fork, and clean Gertrude task workspaces. Load when working on Gertrude, creating, reviewing, or forking GT tasks, choosing an isolated agent workspace, or explaining how to use `gt`.
Create compact, dark-mode, browser-viewable HTML artifacts for specs, plans, PRDs, RFCs, roadmaps, implementation plans, architecture proposals, decision docs, and strategy briefs. Use when the user asks to spec something out, make a plan, draft a proposal, compare options, or produce substantial planning content; save a scannable artifact and return its path instead of dumping a long Markdown answer in chat. Prefer simple disposable docs with visual sketches or UI examples when helpful.
| name | session-history |
| description | Find, inspect, summarize, or search prior pi sessions/conversations. Use when the user asks to refer to previous sessions, inspect session history for a working directory, search across all conversations, recover prior tool output, or audit what was stored. |
Pi stores local session transcripts as JSONL files. Treat them as sensitive: they can contain user messages, assistant messages, tool calls, tool arguments, tool output, bash output, images as base64, cwd/session metadata, branching, compaction summaries, labels, model changes, token usage, and extension messages/state.
Default root:
session_root="${PI_CODING_AGENT_SESSION_DIR:-${PI_CODING_AGENT_DIR:-$HOME/.pi/agent}/sessions}"
Sessions are usually under:
~/.pi/agent/sessions/--<cwd with / replaced by ->--/*.jsonl
Example for /Users/miciah/.dotfiles:
~/.pi/agent/sessions/--Users-miciah-.dotfiles--/*.jsonl
List recent sessions across all working dirs:
find "$session_root" -type f -name '*.jsonl' -print0 | xargs -0 ls -lt | head -50
List sessions for the current working directory:
encoded_cwd="$(pwd | sed 's#/#-#g')"
find "$session_root/--$encoded_cwd--" -type f -name '*.jsonl' -print 2>/dev/null
Search all sessions for text:
rg -n "SEARCH_TEXT" "$session_root"
Show basic metadata for every session:
find "$session_root" -type f -name '*.jsonl' -print0 |
while IFS= read -r -d '' file; do
python3 -c 'import json,sys; h=json.loads(open(sys.argv[1]).readline()); print(f"{h.get(\"timestamp\",\"\")}\t{h.get(\"cwd\",\"\")}\t{h.get(\"id\",\"\")}\t{sys.argv[1]}")' "$file"
done | sort
{"type":"session", ...} with id, timestamp, cwd, version, and sometimes parentSession.{"type":"message", "message": ...}.message.role may be user, assistant, toolResult, bashExecution, custom, branchSummary, or compactionSummary.text, thinking, and toolCall blocks. Tool calls include name and full arguments.isError, toolName, and optional details.id / parentId; do not assume simple linear history if branches exist.When answering from session history, summarize relevant findings and cite session file paths/timestamps. Do not dump large transcripts or secrets unless the user explicitly asks.