Luigit
repositories / pi-ext

pi-ext

bugabingas pi extensions

owned by admin

.system/plans/PX-PLAN-P4FMHCDS-peer-review-skill-delivery/index.md

Raw
Rendered preview

id: PX-PLAN-P4FMHCDS type: plan title: Peer Review Skill Delivery spec: PX-SPEC-O6OLUJJO status: draft

Outcome

Ship a discoverable peer-review skill through the existing PiPi package, independently of Relop, Hornets, Ask, and Dilemma. Implements Peer Review Workflow. Plan prerequisites: none. Relop retirement consumes the usable review route as an integration prerequisite, not as a runtime dependency.

Structure and behavior dependencies

Every behavior uses existing general delegation; optional transport never becomes required.

flowchart TB
  I["Explicit reviewer preference"] --> S["Peer-review skill"]
  S --> P["Fresh PiPi task"]
  P --> E["Investigative evidence and findings"]
  E --> F["Executor evaluates and fixes"]
  F --> C["Continue peer with revised evidence"]
  C --> E
  H["Hornets when available"] -.-> S

Ordered behaviors

  1. Request a useful independent review through PiPi. Add extensions/pipi/skills/peer-review/SKILL.md, following existing pipi-workflows discovery and delegation guidance rather than copying its mechanics. Use explicit task-level model choice, existing deliberately applicable preferences, and existing presets only when they fit; do not add a fixed reviewer preset, pool, or alias registry. Require a substantially different reviewer model and prefer another provider; report known identity and ask if suitable preference is unknown. Grant relevant repository inspection and authorized focused verification, including callers, dependency docs, and actual renderings where needed; no author-selected snapshot ceiling. Keep model availability checks distinct from preference and do not manufacture read-only claims from a shell-enabled tool list. Verify a harmless multi-file defect with an omitted caller and a large relevant file; record whether the peer found the actual consequence, not just whether a delegation completed.

  2. Evaluate findings and recheck actual fixes. The same skill defines outcome, constraints, work/revision, evidence labels, concrete findings, and explicit gaps. Executor verifies consequential findings and performs fixes outside the reviewer task; continue that reviewer session with revised work and verification. Exercise an actual corrected defect, an unsupported allegation, insufficient context, explicit cancellation, and a task with no suitable preferred reviewer. Keep static inference, reproduced results, rendered evidence, and author-reported checks distinct; no persisted review-loop subsystem or automatic delivery action.

  3. Discover the skill outside the source checkout. Use existing extensions/pipi/package.json skills/resources declarations; adjust only if actual packaging requires it. Extend scripts/build/build.test.ts resource-copy and relocated DefaultResourceLoader discovery assertions to include the new skill. Add a concise discovery pointer in pipi-workflows and PiPi README without embedding review procedure in unrelated prompts. Verify the relocated skill resolves its own references and works without loading Relop, Ask, Dilemma, or Hornets. No edits to personal preset libraries, provider credentials, or user settings are part of delivery.

Boundaries

Owned surfaces: extensions/pipi/skills/peer-review/, existing PiPi skill discovery/docs, scripts/build/build.test.ts. Review-specific guidance belongs in the skill; PiPi runtime, model tiers, Dilemma policy, and transport lifecycle remain unchanged. Hornets parity may verify use of another available transport later; absence must not block this plan. Do not claim reliable future model judgment from deterministic resource or scripted-provider tests.

Verification

  • Skill checker plus existing focused PiPi/build tasks prove frontmatter, references, distribution, and native discovery.
  • Harmless evaluation artifacts live in OS temp; preserve RED/GREEN observations and remove owned scratch afterward.
  • Live semantic evaluation requires a deliberately selected different reviewer model; missing preference/access is an explicit blocker, not a reason to use another authenticated provider.
  • Use existing //extensions/pipi:test, lint, verify-package and root //:test scripts/build/build.test.ts as relevant; no implicit full CI or browser E2E.
  • Record source/relocation checks separately from observed review usefulness and minimum-Pi claims.

Exit criteria

  • Root-distributed skill loads from relocated PiPi resources without the optional extensions.
  • A deliberately diverse peer can investigate beyond author-supplied files and report the known defect with consequential evidence.
  • Re-review checks revised evidence while executor retains fixes and delivery authority.
  • Unknown preference, unavailable peer, incomplete investigation, and cancellation are not represented as completed review.
  • Guidance introduces no reviewer pool, identity-maintenance subsystem, or new orchestration engine.
---
id: PX-PLAN-P4FMHCDS
type: plan
title: Peer Review Skill Delivery
spec: PX-SPEC-O6OLUJJO
status: draft
---

## Outcome

Ship a discoverable peer-review skill through the existing PiPi package, independently of Relop, Hornets, Ask, and Dilemma.
Implements [Peer Review Workflow](../../specs/PX-SPEC-O6OLUJJO-peer-review-workflow/index.md).
Plan prerequisites: none.
[Relop retirement](../PX-PLAN-2_9FRWV0-relop-runtime-retirement/index.md) consumes the usable review route as an integration prerequisite, not as a runtime dependency.

## Structure and behavior dependencies

Every behavior uses existing general delegation; optional transport never becomes required.

```mermaid
flowchart TB
  I["Explicit reviewer preference"] --> S["Peer-review skill"]
  S --> P["Fresh PiPi task"]
  P --> E["Investigative evidence and findings"]
  E --> F["Executor evaluates and fixes"]
  F --> C["Continue peer with revised evidence"]
  C --> E
  H["Hornets when available"] -.-> S
```

## Ordered behaviors

1. **Request a useful independent review through PiPi.**
   Add `extensions/pipi/skills/peer-review/SKILL.md`, following existing `pipi-workflows` discovery and delegation guidance rather than copying its mechanics.
   Use explicit task-level model choice, existing deliberately applicable preferences, and existing presets only when they fit; do not add a fixed reviewer preset, pool, or alias registry.
   Require a substantially different reviewer model and prefer another provider; report known identity and ask if suitable preference is unknown.
   Grant relevant repository inspection and authorized focused verification, including callers, dependency docs, and actual renderings where needed; no author-selected snapshot ceiling.
   Keep model availability checks distinct from preference and do not manufacture read-only claims from a shell-enabled tool list.
   Verify a harmless multi-file defect with an omitted caller and a large relevant file; record whether the peer found the actual consequence, not just whether a delegation completed.

2. **Evaluate findings and recheck actual fixes.**
   The same skill defines outcome, constraints, work/revision, evidence labels, concrete findings, and explicit gaps.
   Executor verifies consequential findings and performs fixes outside the reviewer task; continue that reviewer session with revised work and verification.
   Exercise an actual corrected defect, an unsupported allegation, insufficient context, explicit cancellation, and a task with no suitable preferred reviewer.
   Keep static inference, reproduced results, rendered evidence, and author-reported checks distinct; no persisted review-loop subsystem or automatic delivery action.

3. **Discover the skill outside the source checkout.**
   Use existing `extensions/pipi/package.json` skills/resources declarations; adjust only if actual packaging requires it.
   Extend `scripts/build/build.test.ts` resource-copy and relocated `DefaultResourceLoader` discovery assertions to include the new skill.
   Add a concise discovery pointer in `pipi-workflows` and PiPi README without embedding review procedure in unrelated prompts.
   Verify the relocated skill resolves its own references and works without loading Relop, Ask, Dilemma, or Hornets.
   No edits to personal preset libraries, provider credentials, or user settings are part of delivery.

## Boundaries

Owned surfaces: `extensions/pipi/skills/peer-review/`, existing PiPi skill discovery/docs, `scripts/build/build.test.ts`.
Review-specific guidance belongs in the skill; PiPi runtime, model tiers, Dilemma policy, and transport lifecycle remain unchanged.
[Hornets parity](../PX-PLAN-2W7CNSCD-hornets-codex-parity-and-optional-bridges/index.md) may verify use of another available transport later; absence must not block this plan.
Do not claim reliable future model judgment from deterministic resource or scripted-provider tests.

## Verification

- Skill checker plus existing focused PiPi/build tasks prove frontmatter, references, distribution, and native discovery.
- Harmless evaluation artifacts live in OS temp; preserve RED/GREEN observations and remove owned scratch afterward.
- Live semantic evaluation requires a deliberately selected different reviewer model; missing preference/access is an explicit blocker, not a reason to use another authenticated provider.
- Use existing `//extensions/pipi:test`, `lint`, `verify-package` and root `//:test scripts/build/build.test.ts` as relevant; no implicit full CI or browser E2E.
- Record source/relocation checks separately from observed review usefulness and minimum-Pi claims.

## Exit criteria

- [ ] Root-distributed skill loads from relocated PiPi resources without the optional extensions.
- [ ] A deliberately diverse peer can investigate beyond author-supplied files and report the known defect with consequential evidence.
- [ ] Re-review checks revised evidence while executor retains fixes and delivery authority.
- [ ] Unknown preference, unavailable peer, incomplete investigation, and cancellation are not represented as completed review.
- [ ] Guidance introduces no reviewer pool, identity-maintenance subsystem, or new orchestration engine.