feat: Adds stability domain - /memory-leak, /race-condition-repro skills

[MajorLift]

Overview

Adds a new stability domain — a leak kills the tab: availability, not throughput — with two skills.

memory-leak — a two-phase retention review. Phase 1 is static, from the diff: name each retention primitive the change introduces — listener, timer, pending-map entry, subscription, module singleton, growing collection — as a holder → held set → outlived boundary, and pair every acquire against its release. Findings are scoped to what the diff adds. Phase 2 — heap snapshots, retainer graph, a falsifying lifecycle test — is reached only where Phase 1 cannot pair a primitive.

race-condition-repro — forces the interleaving instead of waiting for it: fake timers advanced into the pending window. The proof obligation is that the interleaving happened, not that the assertion passed.

Motivation

A vacuous green that looks exactly like a real one is the shared failure: a missing release is an absence, so a passing suite is what the leak looks like, and a test whose operations ran in sequence exercised no race.

Showcase

Two merged metamask-extension PRs, opposite answers:

PR	Phase 1	Phase 2
#40684	every acquire pairs — no unreleased retention in this diff	not needed
#44352	nothing to pair — the leak is a native object's lifecycle, not the diff's	+~105 MB retained per popup cycle, linear, surviving a forced GC

The slope, not one reading, separates a leak from a working set.

Trial runs on PRs nobody flagged:

PR	Finding
#42823	Gap — listener registered, never removed; charged as latent
#45035	Clean — 2 primitives introduced, both paired

race-condition-repro has no trial run yet, so nothing here speaks to it.

The evidence showcase carries the concurrency example: #44003 (deterministic interleaving) — 38/38 at head, 13/38 failing with the implementation reverted, all inside the retry blocks and 0 elsewhere, on an identical test file.

[MajorLift]

Context budget

What this PR costs an agent, measured from an install rather than read from the diff. Three tiers, and only the first is unavoidable.

Skill	Frontmatter	Selected	+ refs & knowledge
memory-leak	992 chars	~2,701 tok	~5,225 tok

Frontmatter is the only tier paid unconditionally — every agent loads it on every run once the skill is installed, used or not, because it is what the agent reads to decide relevance. The 28 skills across the eleven open skill PRs sit at a median of ~1,716 tokens selected and ~1,860 with references followed. All are within the 1,536-character description budget.

Selected is paid only when the agent picks the skill. + refs & knowledge is the ceiling if every bundled reference is then read; it is a worst case, not an expectation.

Method

tools/install --repo metamask-extension --maturity experimental against this branch at affac8f1f, measured per installed skill directory. Repo overlays are merged into the emitted SKILL.md, so they land in the selected tier rather than being missed by a source-byte count. Token figures are bytes/4 — a proxy for scale, not accounting.

These figures are pinned to the commit above and drift on every push; #96 tracks automating them.