Menü
Consolidate session learning into permanent architecture — extract patterns into skills, instructions, prompts, or memory
| name | meditation |
| description | Consolidate session learning into permanent architecture — extract patterns into skills, instructions, prompts, or memory |
| lastReviewed | "2026-05-13T00:00:00.000Z" |
Transform session insights into durable knowledge. Most sessions don't need it; some have a pattern worth keeping.
/meditateSkip when: the session was routine execution of patterns already encoded. Meditation on every session produces noise; the discipline is to write only what's new and portable.
Scan the session honestly:
Separate signal from noise. For each candidate pattern, ask: "Is this already covered by an existing skill, instruction, or memory?" If yes, skip. If no, route by type:
| If pattern is... | Create / update |
|---|---|
| Reusable domain knowledge | Skill (.github/skills/<name>/SKILL.md) |
| Always-on behavior or rule | Instruction (.github/instructions/<name>.instructions.md) |
| Repeatable workflow / slash command | Prompt (.github/prompts/<name>.prompt.md) |
| Automatable mechanical task (skill-owned) | Skill script (.github/skills/<name>/scripts/*.cjs) |
| Automatable mechanical task (cross-cutting) | Repo script (scripts/<name>.cjs) |
| User preference (cross-project) | User memory (/memories/<name>.md) |
| Project / repo convention | Repo memory (/memories/repo/<name>.md) |
| Cross-session handoff (next session needs to know) | Repo file (HANDOFF.md at repo root) — NOT session memory |
Each artifact gets correct frontmatter, concrete examples (not abstractions), and tables with real data. Avoid the "capabilities list" anti-pattern — describe behavior, not features.
For skills and instructions: include a Trigger or When to fire section so future sessions know when the pattern applies.
If the session arc was substantial, write to .github/episodic/meditation-YYYY-MM-DD-<topic>.md:
# Meditation: <Topic>
**Date**: YYYY-MM-DD
**Focus**: What we worked on
## Accomplished
- [Key outcomes]
## Patterns Extracted
- [What became skills / instructions / memory]
## Lessons
- [Insights worth remembering]
## Open Questions
- [What remains unresolved]
Skip the chronicle for short sessions or when nothing new emerged.
If the user is closing the thread, write to repo-root HANDOFF.md:
# Session Handoff
Last updated: YYYY-MM-DD HH:MM
## Just shipped
- [SHAs / files / outcomes]
## In progress
- [Specific next step + file paths]
## Pending queue
- [ ] [Ordered todos]
## Resume point
- [Where to pick up]
A meditation is complete when:
The five-step protocol above is per-session. The Mall brain also retrains on a longer cadence against two evidence streams:
lastReviewed dates and the weekly catalog-refresh PR signals.| Cadence | Action |
|---|---|
| Weekly | Review the catalog-refresh PR. Note any friction; if a brain artifact mis-routed or under-described, hotfix in the same PR. |
| Monthly | Run currency-audit against any artifact with lastReviewed older than 90 days. Refresh or sunset. |
| Quarterly | Walk every brain artifact. Confirm falsifiability-deadlines haven't fired without action. Sunset what no longer earns its tokens. |
Retraining is critical thinking applied to the brain itself. Each cycle treats the current brain as a hypothesis and tests it against accumulated evidence. Skipping a cycle is the same discipline violation as skipping act-pass on a behaviour-class commit.
| Anti-pattern | Correction |
|---|---|
| Writing a meditation note for every session | Most sessions are routine execution; only write when something new emerged |
| Duplicating an existing skill / memory under a new name | Always grep first: is this already covered? |
| Aspirational notes ("we should do X someday") | Memory is for what was learned, not what was wished |
| Long prose chronicles when a one-line memory suffices | Match artifact size to insight size |
| Skipping the duplication check to "just capture it" | Adds noise that the next session has to filter |
How the Plugin Mall scans, scores, and prunes itself. The Mall is a self-curating repo; this skill is its operational playbook for the weekly catalog refresh and trust scoring. Use when working on the scan pipeline, debugging a stuck weekly PR, or onboarding to Mall internals.
Maintain the source registry in sources/supported-stores.json — add a new third-party plugin store, retire one, refresh metadata, validate the schema. Use when proposing a registry change, after the weekly cron flags a source as unhealthy, or when a candidate store needs evaluation before adding.
Detect, classify, and prune stale source stores in the Mall — define what stale means and how to remove gracefully without breaking downstream consumers.
Evaluate a proposed store for inclusion in Alex_ACT_Plugin_Mall using a quality scorecard
Comprehensive brain file review — external freshness, internal consistency, semantic accuracy — stamp only after full assessment
This skill should be used when building agent evaluation systems: deterministic checks, regression suites, multi-dimensional rubrics, quality gates, production monitoring, baseline comparison, and outcome measurement for agent pipelines.