[FEATURE] Add Pi coding agent as an APM target

[abhishekbiyala]

Is your feature request related to a problem? Please describe.

The Pi coding agent (@earendil-works/pi-coding-agent) is an actively-used harness that APM currently has no target for. Pi users therefore have no manifest/lockfile/upgrade story for their agent context (skills, prompt templates) — the exact gap APM exists to close for every other harness (Copilot, Claude, Cursor, OpenCode, Codex, Gemini, Windsurf, Kiro, Grok, Antigravity).

Describe the solution you'd like

Add pi as a first-class APM target, following the existing opencode/kiro/codex pattern — one TARGET_CAPABILITIES entry plus one KNOWN_TARGETS profile, no new classes.

Pi's loadable project resources (verified against the shipped dist/core/resource-loader.js and dist/config.js):

• skills → converge on the cross-tool .agents/skills/<name>/SKILL.md (Pi natively reads both .pi/skills/ and .agents/skills/); user scope ~/.pi/agent/skills/.
• commands → Pi prompt templates (/name slash commands) → .pi/prompts/<name>.md, reusing the shared claude_command transformer (Pi's description/argument-hint frontmatter and $1/$@/$ARGUMENTS substitution match Claude commands).
• instructions → compile-only (Pi reads AGENTS.md → compile_family="agents").
• No agents primitive — Pi has no native agent-definition directory (sub-agents are an extension capability), so no frontmatter transform is required.
• No hooks. MCP wiring deferred to a follow-up.

Project root .pi/, user scope ~/.pi/agent/, auto-detected on a .pi/ directory, included in --target all.

Describe alternatives you've considered

• Explicit-only target (like antigravity/grok-cloud): rejected — Pi is a mainstream harness with a stable .pi/ config dir, so first-class auto-detection is appropriate.
• A dedicated Pi agent-frontmatter transformer (like OpenCode's): unnecessary, because Pi deploys no agent files.

Additional context

Implementation is ready and opened as PR #2907 (skills + prompt-templates target, tests mirroring kiro/opencode, docs updated, all unit + integration tests green). This issue tracks the feature request that PR resolves.

[github-actions]

Triage decision

accept

Proposed labels

theme/portability
area/multi-target
type/feature
status/in-flight

Milestone

null

Suggested next action

Review and merge PR #2907, which this issue states already implements the target (skills + prompt templates, tests mirroring kiro/opencode, docs updated).

Suggested issue comment

Adding Pi as a first-class target following the existing opencode/kiro/codex pattern (one `TARGET_CAPABILITIES` entry, one `KNOWN_TARGETS` profile, no new classes) is a clean, well-scoped fit for APM's multi-target model, and the resource mapping (skills to `.agents/skills/`, prompt templates via the shared `claude_command` transformer, compile-only instructions, no agents/hooks) is grounded in the actual shipped `resource-loader.js`/`config.js` behavior rather than guesswork.

Since PR #2907 is already open with the implementation, tests, and docs, this is accepted and marked `status/in-flight` rather than `status/accepted` -- the work to review is the PR itself, not a fresh implementation task.

Per-lens notes (collapsed)

DevX UX Expert -- User-Need Reviewer

Adding a new target follows the established, proven pattern (reuse claude_command transformer, no new frontmatter transform needed) rather than inventing new abstractions -- low risk of introducing an inconsistent target surface. First-class auto-detection on .pi/ is the right call for a mainstream harness per the issue's own reasoning against the explicit-only alternative.

Supply Chain Security Expert -- Risk-Surface Reviewer

No new risk surface -- this reuses existing deployment transformers and directory conventions already vetted for other targets (kiro, opencode). No risk-surface implication.

OSS Growth Hacker -- Contributor-Tone Reviewer

Not activated -- author has an already-open implementation PR (#2907) referenced in the same issue, indicating an established contributor rather than a first-time one.

Python Architect -- Architecture Reviewer

Confirmed low cross-cutting impact: the issue explicitly follows the existing TARGET_CAPABILITIES/KNOWN_TARGETS extension pattern with no new classes, reusing the shared claude_command transformer. This is the well-trodden "add a target" path, not a new abstraction -- does not need status/needs-design.

Doc Writer -- Documentation Reviewer

Issue states docs are already updated in PR #2907; the multi-target reference and target-list documentation should be checked for completeness during PR review, but no additional doc-writing task is implied by the issue itself.

APM CEO -- Triage Arbiter

Broadening target support strengthens APM's "works with every harness" positioning at essentially zero design risk given the established extension pattern. Since implementation is already in flight via PR #2907, this tracks as status/in-flight rather than a fresh backlog item -- the actionable next step is PR review.

{
  "decision": "accept",
  "decision_detail": "",
  "theme": "theme/portability",
  "areas": ["area/multi-target"],
  "type": "type/feature",
  "status": "status/in-flight",
  "priority": null,
  "preserved_labels": [],
  "milestone": null,
  "next_action": "Review and merge PR #2907, which this issue states already implements the target.",
  "comment_markdown": "Adding Pi as a first-class target following the existing opencode/kiro/codex pattern is a clean, well-scoped fit for APM's multi-target model, and the resource mapping is grounded in the actual shipped resource-loader.js/config.js behavior rather than guesswork.\n\nSince PR #2907 is already open with the implementation, tests, and docs, this is accepted and marked status/in-flight rather than status/accepted -- the work to review is the PR itself, not a fresh implementation task."
}

---

Triage status: agentic proposal pending human ratification.
Silence is approval. Maintainers can:

• Override any label or milestone above by editing it directly -- human edits are authoritative and will not be reverted on subsequent runs.
• Re-trigger triage by applying the status/needs-triage label, or by removing status/triaged to enroll the issue in the next daily sweep.

Posted by the Triage Panel workflow. See packages/apm-triage-panel for the panel skill.

Generated by Triage Panel · copilot · auto · 120.9 AIC · ⌖ 26.4 AIC · ⊞ 14.1K · ◷