ingest

A command for adding information from a file or web address to an LLM-maintained wiki. It extracts concepts, entities, workflows, and claims, then creates or updates linked wiki pages with citations.

In plain words
What is it for?
Ingesting transcripts, articles, tweets, gists, or repository notes into a wiki and updating existing pages when newer information replaces older claims.
Why use it?
It turns unstructured source material into organized, cross-linked wiki content while requiring claims to be traceable to sources.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/capitalone/context-specs/ingest
Clone the repo
git clone --depth 1 https://github.com/capitalone/context-specs
Per session 39 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,826 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00039 $0.01826
Opus 5 $0.00019 $0.00913
Sonnet 5 $0.00008 $0.00365
Haiku 4.5 $0.00004 $0.00183

Measured 2d ago against content hash df10be6fffc1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

ingest 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 2d 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.

skills/human-loop/wiki-init/assets/commands/ingest.md · 178 lines

How it starts

The opening of the file, as written. The whole thing — 178 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/ingest

You are extending an LLM-maintained wiki at the project root. Read CLAUDE.md first for the layered structure, page taxonomy, frontmatter contract, and hard rules. You must obey those rules. In particular: never modify raw/ (except to save a URL fetch), every wiki claim must cite a [[source]], prefer updating existing pages over creating near-duplicates, and rewrite (don't just append) when a new source supersedes an old claim.

Arguments: $ARGUMENTS

Parse from $ARGUMENTS:

  • Required: a file path under raw/ or a URL.
  • Optional: --type=transcript|article|tweet|gist|repo-notes (auto-detect from path/URL/content if absent).
  • Optional: --yolo — skip the plan-pass approval step.

If $ARGUMENTS is empty or unparseable, ask the user for the source path/URL and stop.

Step 1 — Resolve the source into raw/

Two cases:

A. File path: must already live under raw/. If it's a path outside raw/, refuse and tell the user to move it under the right raw/ subfolder first (transcripts/articles/tweets/gists/repos).

B. URL: fetch the content (use WebFetch). Pick a destination based on --type or the URL shape:

  • gist.github.com → raw/gists/<slug>.md
  • twitter.com / x.com → raw/tweets/<slug>.md
  • otherwise → raw/articles/<slug>.md

The slug is a kebab-case derivative of the title or URL path. Save the fetched content with this frontmatter at the top:

---
source_url: <full URL>
fetched_at: <YYYY-MM-DD>
type: <transcript|article|tweet|gist|repo-notes>
---

This is the only time you write under raw/.

Step 2 — Read the source fully

Use Read on the resolved raw/... path. Read the whole document (use multiple Read calls if necessary; do not truncate).

Step 3 — Extract

Build an internal scratch list. For each item, decide which page type it belongs to:

  • Concepts: ideas, techniques, distinctions, named patterns. Each becomes (or updates) a wiki/concepts/<slug>.md.
  • Entities: people, orgs, tools, products, repos, protocols, regulations. Each becomes (or updates) a wiki/entities/<slug>.md.
  • Workflows: practices with steps, multi-step procedures. Each becomes (or updates) a wiki/workflows/<slug>.md.
  • Claims: specific factual or evaluative assertions. Each will appear in the source-summary's "claims extracted" list and get folded into the relevant concept/entity/workflow page.
  • Open questions: things the source raises but doesn't answer. These go in the "Open questions" section of relevant pages.

Read the full file on GitHub · 178 lines

Changes

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.

  1. 2d ago First seen · 178 lines · 39 tokens per session scan A df10be6fffc1

Subscribe to this mod's changes

ingest is a command published in the GitHub repository capitalone/context-specs (41 stars, last pushed 8d ago), licensed Apache-2.0. It adds 39 tokens to every session and 1,826 once invoked, about $0.0002 per session on Opus 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.