pi-ext

---
id: PX-RESEARCH-3B89B99C
type: research
title: Retired Extension Experiments
---

## Status boundary

This bundle records historical design evidence, not current requirements or implementation commitments.
The current root README and `extensions/` tree list no `goldfish`, `lazy`, `goal`, `shame`, `cmprss`, `overkill`, `tune`, or `leaks` extension.
Legacy documents that say “shipped” or “reconciled” describe their former tree and must not be read as proof of a current extension.
Current extension behavior must be established from current source and its README rather than inferred from these records.

## Durable lessons

A Pi-native extension should use Pi lifecycle, session, resource, UI, and tool seams before introducing a host wrapper, server, proxy, shell language, dashboard, or portability layer.
`cmprss` rejected Headroom’s proxy, native, ML, and provider-wrapper stack in favor of deterministic in-process compression with explicit exact-byte retrieval.
`lazy` retained a small value ladder, canonical idempotent prompt injection, direct deterministic commands, and branch-aware session entries while rejecting multi-host adapters and legacy Ponytail compatibility.
`goldfish` established that task-gated tools need an explicit replayable state machine, visible state, and hard enforcement rather than prompt-only guidance.
`goal` established that a completion judge must be the single atomic transition gate, while durable status must not depend on an optional widget and UI-only messages must stay out of model context.
`shame` favored inference from bounded local evidence and side-effect-free completion over a large typed target surface.
`tune` showed that fixed JavaScript benchmark modules plus direct argument-array process calls are portable, while arbitrary shell command strings create needless platform divergence.
`overkill` separated candidate evaluation from adoption through isolated sessions, frozen model settings, read-only target tools, bounded evidence, atomic checkpoints, strict validation improvement, held-out reporting, and private staging.
`overkill` session mining treated raw session data as sensitive operational material requiring explicit human confirmation, bounded immutable digests, disclosed model transfer, and no autonomous target selection or adoption.
`leaks` favored precision and reversible session safety over maximum detection recall, so ambiguous values remain visible and placeholders require durable authenticated branch-local mappings before tool hydration.
`leaks` also made exact spans, negative corpora, property tests, real agent-loop tests, and reproducible fuzz failures stronger evidence than aggregate redaction counts or direct-hook tests.

## Reuse boundary

These lessons are selection criteria for future work, not a mandate to revive any named extension or its commands, schemas, files, dependencies, prompts, mockups, benchmarks, or checklists.
A successor should retain only independently verified constraints appropriate to its current Pi API and repository architecture.

## Legacy provenance

Historical source: `docs/super/plans/2026-06-22-goldfish.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/plans/2026-07-03-lazy-pi-extension.md` and `docs/super/specs/2026-07-03-ponytail-pi-extension-design.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/plans/2026-07-10-goal-extension-megaplan.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/plans/2026-07-11-shame-command.md` and `docs/super/specs/2026-07-11-shame-command-design.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/specs/2026-07-03-cmprss-pi-extension-design.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/plans/2026-07-16-tune-cross-platform-runner.md` and `docs/super/specs/2026-07-16-tune-cross-platform-runner-design.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/roadmaps/2026-07-13-overkill-roadmap.md`, `docs/super/specs/2026-07-13-overkill-skill-training-design.md`, and `docs/super/specs/2026-07-13-overkill-aftermath-design.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/plans/2026-07-15-overkill-phase-1-contracts-run-state.md`, `docs/super/plans/2026-07-15-overkill-phase-2-isolated-baseline-rollouts.md`, `docs/super/plans/2026-07-15-overkill-phase-3-optimizer-patches-strict-gate.md`, `docs/super/plans/2026-07-15-overkill-phase-4-modes-review-cancellation.md`, `docs/super/plans/2026-07-15-overkill-phase-5-aftermath-harvest-mining.md`, and `docs/super/plans/2026-07-15-overkill-phase-6-aftermath-scoring-integration.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.
Historical source: `docs/super/roadmaps/2026-07-17-leaks-detector-reliability-roadmap.md`, `docs/super/specs/2026-07-17-leaks-detector-reliability-design.md`, `docs/super/plans/2026-07-17-leaks-detector-reliability-phase-1-detector-precision.md`, `docs/super/plans/2026-07-17-leaks-detector-reliability-phase-2-durable-lifecycle.md`, and `docs/super/plans/2026-07-17-leaks-detector-reliability-phase-3-integration-ci.md`, recoverable from Git commit `88660125c313dd5eff4f8c4dee431caa8e0ddd22`.