Context Kit Skills
A family of six Claude skills for building and
maintaining a context kit -- a CLAUDE.md entry point plus a structured
docs/ tree (runbooks, changelog, history, status, per-session notes) that
gives an AI agent durable, low-noise project memory across sessions.
The skills are stack-agnostic and name-agnostic: they describe conventions and
procedures, not any one project. Drop the skills/ directory into a
skills-capable Claude environment and invoke a skill by name.
The skills
| Skill | What it does |
|---|---|
| context-kit-capture | Capture a finished workflow or troubleshooting session as a self-contained task file and update CLAUDE.md. |
| context-kit-check | Read-only integrity, wiring, and hygiene audit of a context kit: version/changelog match, link resolution, orphaned docs, committed-secret and leaked-path scan, date and ordering checks. Includes scripts/check_kit.py. |
| context-kit-ingest | Repeatably fold pre-existing material (an old chat export, a handover, a ticket export) into the kit -- distilled to signal, never stored as a verbatim transcript. |
| context-kit-scaffold | Mine an existing project's docs, conventions, and runbooks into a portable, generic starter kit for the next project. |
| context-kit-split | Reorganize docs to cut per-session context load: move accumulated history out of session-loaded entry-point docs into companion files, leaving a pointer behind. Includes scripts/verify_split.sh. |
| context-kit-steward | Meta-skill that keeps the family coherent: owns the shared conventions charter (references/CONVENTIONS.md) and audits the other skills for drift. |
Conventions they share
The canonical conventions live in
skills/context-kit-steward/references/CONVENTIONS.md:
work-start dating, newest-first ordering, multi-session concurrency safety,
reference-don't-embed, hygiene (no verbatim transcripts, no secrets, no leaked
absolute paths), and skill-authoring limits.
Layout
skills/
context-kit-capture/ SKILL.md
context-kit-check/ SKILL.md + scripts/check_kit.py
context-kit-ingest/ SKILL.md + references/
context-kit-scaffold/ SKILL.md + references/
context-kit-split/ SKILL.md + scripts/verify_split.sh
context-kit-steward/ SKILL.md + references/CONVENTIONS.md
Usage
These are Agent Skills.
Place the contents of skills/ where your Claude environment discovers skills,
then invoke one by name (for example, ask Claude to "check my CLAUDE.md" to
trigger context-kit-check). Each SKILL.md documents its own triggers and
procedure.
License
MIT.
No comments yet
Be the first to share your take.