sf-pi

• Install SF Pi only for the current project with pi install -l git:github.com/salesforce/sf-pi.

• If terminal glyphs appear as ?, run /sf-setup-fonts and select MesloLGM Nerd Font Mono in your terminal.

• Update the three core installations with:

  pi update --self
  pi update git:github.com/salesforce/sf-pi
  npm install --global @salesforce/cli@latest

macOS, Linux, and WSL are the primary targets. Native Windows is supported, with WSL recommended for parity with Unix shell tooling. SF Pi's stable Pi range is currently >=0.82.0 <1.0.0. /sf-pi doctor reports whether the installed Pi and SF Pi versions are current.

The pi runtime itself emits one anonymous install/update version ping to https://pi.dev/api/report-install after a fresh install or changelog-detected update. sf-pi opts you out of this ping by default: on the first session after installing sf-pi, enableInstallTelemetry: false is written to pi's global settings.json if (and only if) the key is currently unset. An explicit user opt-in (true) is always preserved.

The sf-welcome splash shows a Privacy: telemetry off (sf-pi default) row at every launch so the posture is auditable at a glance. Manage it via:

/sf-pi telemetry status     # show the current state and source
/sf-pi telemetry on         # opt back in (writes enableInstallTelemetry: true)
/sf-pi telemetry off        # opt out (writes enableInstallTelemetry: false)

What sf-pi's default does not touch (intentional):

• Latest-version probe to https://pi.dev/api/latest-version stays enabled so users still see security/feature update nudges. Disable separately with PI_SKIP_VERSION_CHECK=1 or master-kill with PI_OFFLINE=1.
• PI_OFFLINE / PI_TELEMETRY env vars or shell rc files — sf-pi never edits your shell environment. The default lives only in pi's settings.json.
• LLM provider traffic — always determined by the provider you configure.

Maintainers archive aggregate GitHub metrics, such as repository views, repository clones, and release download counts, through a scheduled GitHub Actions workflow. These platform metrics help measure discovery and distribution without adding client-side telemetry.

See docs/telemetry.md for the full privacy policy and future telemetry requirements.

The startup splash can show a small Announcements panel for sf-pi maintainer notes and update nudges. Announcements come from the bundled catalog/announcements.json, optionally merge with a hosted JSON feed, and fail silently when offline.

Useful commands and controls:

/sf-pi announcements                  # list active announcements
/sf-pi announcements dismiss <id>     # hide one item
/sf-pi announcements reset            # clear local dismissals
SF_PI_ANNOUNCEMENTS=off pi            # disable the feature for one run
SF_PI_ANNOUNCEMENTS_FEED=off pi       # keep bundled notes, skip remote feed

Persistent opt-out can also live in Pi settings:

{ "sfPi": { "announcements": false } }

Or keep bundled/update notices while disabling only the hosted feed:

{ "sfPi": { "announcements": { "feedEnabled": false } } }

Every slash command lives inside a bundled extension. This table is the fastest way to map a command to the extension that owns it. For subcommands and flags, follow the link into each extension's README, or see the auto-generated docs/commands.md for a richer per-extension view.

Every slash command exposed by a bundled extension. See each extension README for subcommands and flags.

All eight packages are MIT-licensed and install per-user (global scope) by default.

Full manifest with source URLs, license info, and per-item rationale strings: catalog/recommendations.json.

• Open it: /sf-pi recommended
• One-liner install: /sf-pi recommended install <id>
• Whole bundle: /sf-pi recommended install bundle:default
• Decline + forget: pick Never in the checklist or /sf-pi recommended remove <id>

First-run behavior:

• On every session_start, sf-pi checks whether the manifest's revision differs from what you've already acknowledged. If it does, a one-line nudge appears in the footer status (✨ sf-pi: N new recommended …).
• Nothing installs automatically. You stay in control — run /sf-pi recommended when you're ready, pick what you want with Space, press Enter to apply.
• Decisions are sticky: items you installed or declined are never re-prompted across sessions.
• Opt out entirely with SF_PI_RECOMMENDATIONS=off in your environment.

Proposing a new recommendation: see CONTRIBUTING.md.

Pi natively loads skills from ~/.pi/agent/skills/ and ~/.agents/skills/. Skill libraries from other harnesses — Claude Code (~/.claude/skills), OpenAI Codex (~/.codex/skills), and Cursor (~/.cursor/skills) — require a one-line settings edit to load in pi:

// ~/.pi/agent/settings.json
{
  "skills": ["~/.claude/skills", "~/.codex/skills"]
}

/sf-pi skills does this for you. Run it and sf-pi:

1. Scans those three directories on disk, counts the skills it sees in each, and cross-references the list with your current settings.skills[].
2. Opens a checklist — Space toggles a root, Enter applies.
3. Writes the delta back to ~/.pi/agent/settings.json and reloads so the newly wired skills load immediately.

The splash also shows a single-line nudge under Recommended whenever it detects an external root on disk that isn't yet in your settings:

• Interop  2 external skill roots (41 skills detected)
  → /sf-pi skills

Skills work side-by-side across harnesses — wiring a Claude Code directory here does not copy, move, or touch the files in any way. Pi reads them in place and Claude Code continues to use them unchanged.