v2.0.0
Procedural Memory — Skills That Adapt to You
skillpm v2.0.0 introduces a cognitive memory subsystem that makes skill management self-adaptive. Skills strengthen with use, decay with disuse, and adapt to your project context — so agents always see the most relevant skills.
Highlights
- 6-layer architecture: observation → context → scoring → feedback → consolidation → adaptive injection
- 4-factor activation scoring:
Score = 0.35×Recency + 0.25×Frequency + 0.25×ContextMatch + 0.15×Feedback - Context-aware: auto-detects project type (Go, Node, Python, Rust, Java, Ruby), frameworks, and task signals
- Adaptive injection:
skillpm inject --adaptiveinjects only the working-memory subset - 14 new CLI commands under
skillpm memory - Zero new dependencies — built on stdlib + existing go-toml/v2
- CPU < 5% overhead — benchmarked with regression detection
New Commands
skillpm memory enable/disable # toggle the subsystem
skillpm memory observe # record skill usage events
skillpm memory scores # show activation scores
skillpm memory working-set # skills in working memory
skillpm memory explain <skill> # detailed score breakdown
skillpm memory rate <skill> +1/0/-1 # explicit feedback
skillpm memory consolidate # recompute & promote/demote
skillpm memory recommend # archival suggestions
skillpm memory set-adaptive on/off # default adaptive mode
skillpm inject --agent claude --adaptive # smart injectionQuick Start
skillpm memory enable
skillpm memory observe
skillpm memory scores
skillpm inject --agent claude --adaptiveStats
- 7 new internal packages (
internal/memory/*) - +5,400 lines of Go code
- ~131 unit tests across all memory packages (>80% coverage)
- 4 E2E test functions
- Benchmark suite with CI regression detection
Docs
- Procedural Memory Guide — comprehensive 6-layer walkthrough
- CLI Reference — updated with all memory commands
- Config Reference — new
[memory]section