Skill agent session handover
Use this skill when the user asks one local AI agent to inspect, continue from, hand off, summarize, or compare another local agent session, especially Cursor agent transcripts under ~/.cursor/projects, Codex sessions under ~/.codex/sessions, and Claude Code transcripts under ~/.claude/projects.From its SKILL.md
npx -y skills add oguzkr/skill-agent-session-handoverAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 15 days oldThe repository was created 15 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.9 KB, 889 tokens by cl100k_base, as published. Nobody here has run it
Agent Session Handover
Use this skill to read local JSONL session transcripts across agents without pretending there is live chat sync.
Ground Rules
- Be explicit about the source: Cursor transcript, Codex session, Claude Code transcript, or another supplied JSONL path.
- Treat session access as file-based. Do not claim direct access to another app's in-memory conversation.
- Prefer digesting the last relevant messages over dumping whole transcripts.
- Do not write into original transcript files.
- Session transcripts can contain secrets, private paths, and tool output. Summarize only what is needed.
- Check file mtimes when recency matters; a "latest" file can still be stale.
Quick Start
Set the script path:
export AGENTS_SKILLS_ROOT="${AGENTS_SKILLS_ROOT:-$HOME/.agents/skills}"
export SESSION_HANDOVER="$AGENTS_SKILLS_ROOT/agent-session-handover/scripts/agent_session_handover.py"
List recent sessions:
python3 "$SESSION_HANDOVER" list --source all --limit 10
python3 "$SESSION_HANDOVER" list --source cursor --limit 5
python3 "$SESSION_HANDOVER" list --source codex --limit 5
python3 "$SESSION_HANDOVER" list --source claude --limit 5
Read the latest session:
python3 "$SESSION_HANDOVER" latest --source cursor
python3 "$SESSION_HANDOVER" digest "$(python3 "$SESSION_HANDOVER" latest --source codex)" --last 12
python3 "$SESSION_HANDOVER" digest "$(python3 "$SESSION_HANDOVER" latest --source claude)" --last 12
Find a Claude Code session by its title. The list label is the session's
title (the name shown in Claude Code), so match on it and read that path:
python3 "$SESSION_HANDOVER" list --source claude --limit 20
python3 "$SESSION_HANDOVER" list --source claude --json \
| python3 -c "import json,sys; rows=json.load(sys.stdin); print(next(r['path'] for r in rows if 'Conversion optimization' in r['label']))"
Read a known transcript:
python3 "$SESSION_HANDOVER" digest /path/to/session.jsonl --last 12
python3 "$SESSION_HANDOVER" show /path/to/session.jsonl --last 20
Prepare handoff text for another agent:
python3 "$SESSION_HANDOVER" handoff --from cursor --to codex --last 12 --task "Continue this work."
python3 "$SESSION_HANDOVER" handoff --from codex --to cursor --last 12 --task "Review the decision and adjust the Cursor plan."
Workflow
- Locate the session with
listor use the path supplied by the user. - Run
digestfirst. Useshowonly when exact wording matters. - State what file was read and whether it is current enough for the request.
- Continue from the extracted context, or produce
handofftext for the other agent.
Supported Sources
- Cursor main agent transcripts:
~/.cursor/projects/*/agent-transcripts/*/*.jsonl - Cursor subagent transcripts: included only when
--include-subagentsis passed. - Codex sessions:
~/.codex/sessions/**/*.jsonl - Codex archived sessions:
~/.codex/archived_sessions/*.jsonl - Codex thread names: resolved from
~/.codex/session_index.jsonlwhen available. - Claude Code transcripts:
~/.claude/projects/<project-slug>/*.jsonl. The record kind is the top-leveltype(user/assistant); the role ismessage.role. Thelistlabel is the session's custom title, so a user request like "read the 'xxx' session in Claude" maps to the row whose label isxxx.
When To Stop
Ask the user for a specific transcript path if:
- more than one recent session plausibly matches the request,
- the latest session belongs to a different project,
- the transcript format cannot be parsed safely.
What ships with it: 6 files
172.4 KB alongside SKILL.md, 1 of them executable
agents/
- openai.yaml297 B
assets/
scripts/
- agent_session_handover.pyruns16.7 KB
- .gitignore33 B
- LICENSE1.0 KB
- README.md2.4 KB