How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/hu3rror/my-skills --skill pi-session-searchBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
Pi session search
The one command that finds past-session content, replacing ad-hoc grep/python archaeology.
Steps
1. Confirm the sessions root
echo "$PI_SESSION_FILE" # current session; its dirname's parent is the root
ls ~/.pi/agent/sessions # the root (default)
python scripts/pi-session-search.py --root DIR "pattern" # point at a custom rootDone when: you know the root path — ~/.pi/agent/sessions by default, or the --root DIR you pass for a run that used its own --session-dir.
2. Full-text file-level search
python scripts/pi-session-search.py "pattern" # relative to this skill's dirOutput: hit list grouped by repo directory, each file with a hit count and a UTC timestamp spanning the session. Filenames sort newest-last (UTC timestamps prefix them).
Done when: you can name every session file containing the pattern, or state confidently that none do (an empty report means no hit).
3. Structured queries (--json)
--json requires at least one filter (--role / --tool / --field); add a pattern as the last positional ("" matches all).
python scripts/pi-session-search.py --json --role assistant "pattern" # a role
python scripts/pi-session-search.py --json --role assistant --tool subagent "" # tool-call events only
python scripts/pi-session-search.py --json --tool subagent --field "model=" "" # args with a model key (any value)
python scripts/pi-session-search.py --json --tool subagent "sensenova-6.8" # call args containing the text
python scripts/pi-session-search.py --show 5 "pattern" # file mode: 5 snippets per hit fileEach event line prints: timestamp, role, event type, the tool name (for tool calls), and a truncated content snippet.
Done when: your question maps to a named filter, and the output answers it — e.g. "the subagent calls in that session carried model 0/4 times".
4. Drill into a hit
Once you know the file, open it where the hit was:
grep -n "pattern" ~/.pi/agent/sessions/--C--Users-Hue-Repos-<repo>--/<session>.jsonlDone when: you have read the surrounding lines in context.
Reference: environment facts
- Session dirs start with `--` (they encode the cwd, e.g.
--C--Users-Hue-Repos-foo--).ls -d */,for f in $(ls -d */), anddu */20*.jsonlmisparse them as options — use the script, orfind/rgwith an explicit./or--separator. - JSONL, one event per line:
typeissession|message|model_change|thinking_level_change|custom_message| ... A message event's body ismessage.role+message.content[](parts of type text / thinking / toolCall). Tool calls live in assistantcontent[]parts withtype: "toolCall"; tool results are their own events (message.role == "toolResult") carrying the returned text — event-leveldetailsis not written. - Thinking-level evidence has two landing spots: the requested level in a
set-thinking-leveltoolCall'sarguments.level; the applied session level in athinking_level_changeevent (plainthinkingLevelfield). - Python prints Chinese correctly (PYTHONUTF8 active) — no wrapper boilerplate needed.
- Scale: hundreds of JSONL files, ~150 MB total. A full-scan python run is seconds;
rgis faster for pure text. - Output is capped; narrow with filters or
--show Nrather than widening the scan.