← ClaudeAtlas

ilk-self-improvelisted

Plan toolkit improvements from the shared improvement backlog. Reads open candidates emitted by /ilk-feedback, formats a task description, and delegates to /ilk-plan on the ilk-skills repo. The resulting master is auto-gated draft+supervised_only (self-modifying). Triggers: "/ilk-self-improve", "improve ilk", "self-improve ilk", "ilk 自我改进".
inluck-net/ilk-skills · ★ 0 · AI & Automation · score 64
Install: claude install-skill inluck-net/ilk-skills
# ilk-self-improve — backlog → plan adapter A source adapter for the ilk loop, analogous to `/ilk-lark`. It pulls from the **improvement backlog** (populated by `/ilk-feedback` when a postmortem finding is a toolkit/process gap) and hands a formatted task description to `/ilk-plan`. ## When to use - The user says "/ilk-self-improve", "improve ilk", "self-improve ilk", or "ilk 自我改进". - After a `/ilk-feedback` postmortem surfaced toolkit gaps and the user wants to plan fixes. ## Boundary This skill is a **planner**, not an executor. It produces a plan; a human releases and runs it (the master is `draft` + `supervised_only` because it edits the toolkit itself). It does NOT auto-apply changes. ## Workflow 1. Run `build_task.py` to read open candidates from the improvement backlog. 2. If the backlog is empty, report "nothing to improve" and stop. 3. Otherwise, hand the task description to `/ilk-plan` on the ilk-skills repo. 4. `/ilk-plan` auto-gates the resulting master as `draft` + `supervised_only` (self-modifying batch) and auto-registers the project. ## Key files - `scripts/build_task.py` — reads backlog, emits task description. - Backlog store: `~/.ilk-data/ilk-skills-improvements/candidates.json` (managed by `ilk-feedback/scripts/improvement_backlog.py`). ## See also - `/ilk-feedback` — emits upstream candidates into the backlog. - `/ilk-plan` — the planning core this adapter delegates to. - `skills/ilk-feedback/scripts/improvement_backlog.py` — backlog s