Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/tpierrain/kenjakunpx agentmods add skills/tpierrain/kenjaku/consolidateWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tpierrain/kenjaku/consolidate)<a href="https://agentmods.dev/skills/tpierrain/kenjaku/consolidate"><img src="https://agentmods.dev/badge/skills/tpierrain/kenjaku/consolidate/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tpierrain/kenjaku/consolidate"><img src="https://agentmods.dev/badge/skills/tpierrain/kenjaku/consolidate.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00164 | $0.03983 |
| Opus 5.5 | $0.00066 | $0.01593 |
| Sonnet 5.5 | $0.00033 | $0.00797 |
| Haiku 4.5 | $0.00016 | $0.00398 |
Grade A, and why
consolidate scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 222 lines — stays where its author put it; the contents beside it link to each section on GitHub.
consolidate — promote raw captures into the durable wiki ("compile the notes")
A second brain accretes raw capture fast (meetings, daily logs, transcripts) but the higher-order pages — the person, the topic, the project — lag behind: a name gets mentioned ten times and still has no page, a topic page goes stale while fresher notes discuss it. This skill is the "compile" half of the Karpathy LLM-Wiki discipline (see ADR 0033): on demand, it promotes the substance of raw captures into the entity/topic pages and weaves the backlinks, so the wiki (and the RAG's neighbourhood signal) stays current.
Principle
One promise: turn what raw captures already say into durable, well-linked pages — proposed first, written only on yes.
- 🔴 Nearly every gesture here is a genuine question, and that is the right tier (ADR 0043). Creating a page decides how the vault will name a person or a subject from then on, and every later resolution resolves against it; a contradiction between a note and a page is never settled by the brain. A 🔴 is not a failure of the doctrine. What this skill changes is not whether it asks but how often: one message for the whole batch, and each question saying what is at stake before it lists the options. The one 🟡 here is saying what was left for later — a fact, not a request.
- Detection is deterministic (ADR 0009): a pure scanner lists exactly what needs consolidating.
- The merge is judgment (LLM): parallel read-only sub-agents draft each page, never the main context
(the
sync-sourcesfan-out/fan-in shape, to avoid context rot). - The write is deterministic and confirmed (reuse Track B /
/file-back): a new page goes through the builder (conformant by construction, refuses to overwrite); a living page gets a confirmed dated append. So consolidation never re-introduces the defects/lintreports. - Contradictions are flagged, never adjudicated silently (Track D): when a fresher capture CONFLICTS
with a fact the page already states (not merely adds to it), the draft surfaces it as a
### contradictionsentry and the human decides which truth stands. Consolidation folds in additively; it never overwrites a stated fact to resolve a conflict on its own. - Bounded, resumable (stateless), honest: a capture is a candidate purely because it is fresher
than the page it feeds (or that page doesn't exist). Refresh a page and its
updated:moves past the capture, so the candidate drops off on its own — no state file, nothing to seed or corrupt.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed · +32 lines 82edc87d78cb
- 1mo ago Changed b86c62565762
- 1mo ago First seen · 190 lines · 164 tokens per session scan A c139e2ce8e05
consolidate is a skill published in the GitHub repository tpierrain/kenjaku (92 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 164 tokens to every session and 3,983 once invoked, about $0.0007 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
refresh-context
A context snapshot updater for Core Context.md, a summary file built from nine main system files and selected essays. It refreshes that summary when the underlying material changes or the snapshot is more than 30 days old.
brain-query
Semantic two-tier retrieval over the personal brain at /brain/. Queries brainwiki (synthesized pages) first, falls back to brainraw (raw source chunks) if confidence is low, and always reports which layer answered with a confidence score. Use when the user says "/brain-query", "search the brain", "semantic search", or…
wiki-query
Answer a question strictly from the user's brain at /brain/wiki/. Never answer from training memory on topics the wiki covers. Always cite the page used. Use when the user says "/wiki-query", "ask the brain", or asks a factual question about their work, clients, or projects.
obsidian-bases
Explain, draft, and validate Obsidian Bases .base files with filters, formulas, properties, summaries, and table, card, list, or version-gated kanban views. Use for Obsidian Bases, database-like vault views, dynamic tables, reading lists, task trackers, filters, formulas, summaries, and .base file edits.
wiki-fold
Create a bounded, extractive, structurally idempotent rollup of recent Obsidian wiki log entries, with dry-run preview by default and one optional transaction apply. Use for manual log compression without modifying child pages. Triggers: fold the log, run a fold, run wiki-fold, log rollup, roll up log entries, commit…
yopedia
Save research and reflections into (and recall from) your personal knowledge vault (yopedia) — your second brain.