Stats
Actions
Tags
From agent-learning-compounder
Session lifecycle synthesiser. Use this skill when the user asks any variant of: "what's next", "what should I do", "session start", "starting up", "session end", "ending session", "wrap up", "where did I leave off", "where was I", "sum up", "summary", "recap", "context dump", "what was I doing", "next move", "next step", "pick up where I left off", "catch me up", "what do I work on", "hva er neste", "hva skal jeg gjøre", "hvor var jeg", "neste steg", "avslutt sesjon", "oppsummering".
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-learning-compounder:alc-nextThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Thin trigger skill. All synthesis logic lives in `bin/alc_next_action.py`
Thin trigger skill. All synthesis logic lives in bin/alc_next_action.py
(the next_action function) — tested, version-controlled, compoundable.
mcp__alc__next_action with the detected intent and the current
repo path.| User says | intent param |
|---|---|
| "what's next", "what should I do", "next step" | start |
| "session start", "starting up", "booting up" | start |
| "session end", "ending", "wrap up", "close out" | end |
| "where did I leave off", "where was I", "pick up" | leftoff |
| "recap", "summary", "what happened", "sum up" | recap |
| unclear / general | auto |
| Norwegian: "hva er neste", "neste steg" | start |
| Norwegian: "hvor var jeg", "hva holdt jeg på med" | leftoff |
| Norwegian: "oppsummering", "avslutt" | end |
Call: mcp__alc__next_action
Args: { "repo": "<current repo path>", "intent": "<detected intent>" }
Relay the response as follows:
suggested.skill is non-null, offer to run it:
"Suggested:
/<skill><args>" (or just/<skill>if args is null) "<suggested.prompt>"
"Alternatives: / — "
Do not dump the raw JSON. Do not add commentary beyond what the response
provides — the synthesiser's rationale field already explains the
recommendation.
Each call writes <state-root>/repos/<repo-id>/reports/latest-next-action.json.
This is intentional — the dashboard and session-start hook can read the most
recent synthesised answer without re-invoking the MCP tool.
python3 bin/alc_next_action.py directly (not yet wired as a CLI, but
the module is importable).npx claudepluginhub beeard/agent-learning-compounder --plugin agent-learning-compounderCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.