opentraces-pi

OpenTraces capture/search integration for the Pi coding agent

Packages

Package details

extensionskillprompt

Install opentraces-pi from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:opentraces-pi
Package
opentraces-pi
Version
0.1.1
Published
Jun 11, 2026
Downloads
not available
Author
jayfarei
License
MIT
Types
extension, skill, prompt
Size
27.4 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./src/index.ts"
  ],
  "skills": [
    "./skills"
  ],
  "prompts": [
    "./prompts"
  ],
  "image": "https://opentraces.ai/og.png"
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

opentraces-pi

Pi package for OpenTraces. Installs a thin Pi extension, an OpenTraces skill, and prompt templates. The extension observes Pi lifecycle/tool/provider/tree events and forwards normalized sidecars to the Python opentraces bridge.

Install

pi install npm:opentraces-pi

Local development:

pi install ./packages/opentraces-pi
opentraces setup pi --project --local

Package install is quick and reversible. It does not silently install Python, start services, authenticate HuggingFace, enable remote sync, or turn on optional security tools. Capture itself is opt-out: under global tracking (the default) the extension auto-enrolls each repo on first capture, into a private + review-required bucket, once the opentraces CLI is present. Opt out with opentraces config tracking-mode manual or a per-project excluded marker; opentraces init --agent pi still enrolls a repo explicitly.

Raw provider bodies are default-off. Structured context/provider metadata can feed Context Tree; raw bodies are only retained when explicitly opted in and stay local/security-gated.

In Pi

Slash commands are TUI actions for humans:

  • /ot-capture-status — show package, CLI, project capture, raw-body, sidecar, and bucket setup status.
  • /ot-setup — guided local-first setup; asks before enabling project capture.
  • /ot-search <query> — search local retained bucket traces.
  • /ot-trace <trace-id> — resolve one trace and show compact tool evidence.
  • /ot-standup — build a bounded recent-work trace packet.
  • /ot-capsule [trace-id] — preview a trace capsule, or show bucket status without an id.
  • /ot-dataset — list local workflow-built datasets.

Model-facing tools expose the same surfaces: ot_capture_status, ot_search, ot_trace, ot_standup, ot_capsule, and ot_dataset. A typical private bucket workflow is /ot-search farewell helper, then /ot-trace <trace-id> for the selected candidate. ot_search accepts a limit and is read-only against the local snapshot; sync remote traces first with opentraces bucket remote pull. ot_trace can include a compact trace map.

The extension is fail-open. Direct slash commands are not recorded as model tool calls; model-invoked ot_* tools are captured as read-only opentraces_retrieval tool calls. Dataset/capsule actions are retrieval-first; mutating operations such as dataset creation or capsule export should require an explicit user request.