Daily use

After install, kenkeep runs itself. Capture and injection happen on their own; the only thing you ever do by hand is curate, review, and commit.

TIP

Most days you do nothing. When the nudge appears, it’s three steps: /kk-curate, then git diff, then git commit what you want to keep.

The loop

Code with your AI assistant as usual.
When you see the curate nudge (or whenever you feel like it), run /kk-curate.
If the curator reports any contradictions, the skill walks you through each one in-session and applies your chosen resolution.
Inspect the resulting changes under .ai/kenkeep/nodes/ with git diff (or your preferred diff tool, e.g. self-review).
git commit what you want to keep; git restore <path> to discard.

The pre-commit hook regenerates INDEX.md and GRAPH.md and stages them into the same commit, so the injected index never drifts from the committed nodes.

That is the entire daily workflow. The rest of this page is reference: the three skills, and the two manual actions you reach for occasionally.

Skills

Three in-session skills do all the work. Run them inside your harness session; the LLM call happens in that same session, with the model, prompt cache, and tools you already use interactively.

Skill	What it does	When you reach for it
`/kk-add`	Conversationally gathers a node’s fields, checks `INDEX.md` for overlap, writes it under `nodes/`.	Any time you want to capture a fact on the spot.
`/kk-curate`	Reads pending captured sessions, drafts proposed notes under `nodes/`, rebuilds `INDEX.md`/`GRAPH.md`, and walks you through any contradictions with the `y/n/s/k` prompt.	The daily loop, when nudged.
`/kk-bootstrap`	Seeds nodes from your existing docs — a one-time setup step, see Installation → Seed from existing docs.	Once, at setup.

CAUTION

Run only one LLM skill at a time per repo. /kk-curate and /kk-bootstrap are single-author by design and take no cross-process lock, so concurrent runs against the same repo can silently waste work (sessions reprocess on the next run — no data loss). See Architecture → Locking.

The curate nudge

You don’t have to remember to curate. SessionStart counts pending session logs and, once the queue is worth your attention, appends a one-line nudge to the injected context and prints a visible warning to stderr. It escalates to a loud heading when the queue is large or stale. The thresholds are configurable — see Internals → Hooks.

Curate

In a harness session:

/kk-curate

The curator reads every captured session that’s been processed but not yet curated and applies its decisions directly to nodes/:

Decision	Effect
add	Writes `nodes/<kind>/<id>.md`. Fails loud if the file already exists.
modify	Overwrites the target node. Fails loud if `target_node_id` is missing on disk.
contradict	Records the conflict under `.ai/kenkeep/conflicts/<id>.md` with `status: pending`; writes nothing to `nodes/`.
drop	No change.

Fast path

Zero conflicts and zero failures → the skill prints one summary line and exits. Once a project stabilizes, this is the common case.

Conflict walkthrough

When the curator reports contradictions, the skill walks each one with the existing node shown side-by-side and a single-character prompt:

Accept this proposal? [Y/n/s/k] (default: Y)

Key	Action
`y`	Accept: rewrite the node with the proposed body, then `git restore` the conflict file.
`n`	Reject: `git restore` the conflict file; node unchanged.
`s`	Skip: leave the conflict file pending; it re-surfaces next pass.
`k`	Keep: `git commit` the conflict file as a historical record; node unchanged.

Defaults are heuristic per conflict (small diffs default y, rewrites n, otherwise s). Long forms (yes, skip) and uppercase work; anything else is re-prompted.

Add knowledge manually

/kk-add writes a node directly from the current session. Review with git diff and commit. Reach for it when you want to capture a decision the moment you make it, without waiting for the next curate pass.

Status

npx kenkeep status reports queued captures, pending logs, unresolved conflicts, and node counts.