Memoize Skill Capability Classification by Semantic Input

[Trecek]

Summary

Introduce a process-local, thread-safe, weighted LRU around capability-evidence classification in src/autoskillit/workspace/skill_capabilities.py. The cache normalizes the effective logical skill name before lookup and uses the exact (canonical_content, effective_logical_name) pair as its semantic key, preserving classification behavior while reusing immutable evidence tuples for repeated inputs.

Requirements

• Cache exact semantic inputs while ensuring content or effective-name changes classify independently.
• Bound retained state by entry count, aggregate accounted bytes, and a per-input admission ceiling.
• Keep full-document scans outside the cache lock and coordinate concurrent cold callers with generation-scoped single-flight state.
• Preserve existing evidence fields, ordering, source spans, authenticity and capability decisions, projection behavior, and resume authority.
• Keep the implementation dependency-free within the IL-1 workspace layer.

Verification

• task test-check — PASS (15057 passed, 1491 skipped, 27 xfailed)
• Exact regression manifest — PASS (293 passed)
• Cache-focused tests — PASS (26 passed)
• pre-commit run --all-files — all hooks passed
• Representative benchmark — 2,708 classifier calls reduced to 141 underlying scans across 141 unique semantic keys, with an unchanged findings fingerprint
• General paired benchmark — unchanged 23,753 JUnit test cases; median elapsed time reduced from 306.70s to 199.52s

Implementation Plan

Plan file: .autoskillit/temp/make-plan/memoize_skill_capability_classification_plan_2026-07-25_125953.md

🤖 Generated with Claude Code via AutoSkillit

[Trecek]

AutoSkillit PR Review — Verdict: approved_with_comments

[Trecek]

AutoSkillit review: warning-only findings detected. See inline comments — no blocking changes required.

[Trecek]

AutoSkillit PR Review — Verdict: approved_with_comments

[Trecek]

AutoSkillit review: warning-only findings detected. See inline comments — no blocking changes required.