# Okf Consume Bundle

> Use when a user wants Codex to consume, search, read, cite, review, or answer from an existing OKF bundle, especially a GitHub-hosted bundle URL.

- Skill: `thomasrohde/okf-consume-bundle` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add thomasrohde/okf-consume-bundle`
- Raw SKILL.md: https://api.skillmd.com/api/skills/thomasrohde/okf-consume-bundle/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: thomasrohde (https://skillmd.com/u/thomasrohde)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/thomasrohde/okf-consume-bundle

---


# OKF Bundle Consumption Workflow

Use this skill when the user wants to use an existing Open Knowledge Format bundle as context.

## Required Behavior

1. Identify the bundle source: attached alias, local path, GitHub URL, or registry entry.
2. If the user gives a GitHub URL, call `okf_attach_github_bundle` unless the bundle is already attached.
3. If the user gives a local path and wants reuse, call `okf_attach_local_bundle`.
4. Inspect attach diagnostics: source URL, commit SHA when available, bundle path, validation errors, validation warnings, freshness, and concept count.
5. Read `index.md` and `log.md` when available.
6. Use `okf_search_concepts` for the user's question or review topic.
7. Use `okf_read_concept` for high-scoring concepts.
8. Use `okf_related_concepts` when relationships matter.
9. Use `okf_prepare_answer_context` before final answers that claim to be bundle-grounded.
10. Use `okf_bundle_overview` for a broad bundle summary before answering open-ended overview questions.
11. Cite concept IDs or paths after material claims.
12. Distinguish direct bundle facts, inferences from bundle content, and external knowledge.
13. If the bundle is durable and likely to be reused, offer to add a scoped `AGENTS.md` instruction so future Codex sessions consult the bundle for relevant tasks.

## Tool Availability

Prefer MCP tools. If the MCP tools are unavailable in the chat, use the CLI-equivalent commands before direct file lookup:

```bash
python tools/okf_tool.py search <bundle> "<query>"
python tools/okf_tool.py context <bundle> "<question>" --mode strict
python tools/okf_tool.py overview <bundle>
python tools/okf_tool.py generate-chatgpt-usage <bundle> --repo <owner/repo>
python tools/okf_tool.py mcp-diagnostics
```

Use direct file reads only as a last fallback when neither MCP nor CLI access is available.

## Answer Modes

Strict mode:

- Answer only from the OKF bundle.
- Say when the bundle does not contain enough information.
- Do not supplement with general knowledge unless the user asks.

Hybrid mode:

- Use the bundle first.
- Label external knowledge clearly.
- Do not override bundle facts silently.

## Citation Style

Prefer concept path citations:

- `[data/customer-identifier]`
- `[systems/payment-router]`
- `[controls/customer-data-classification]`

For a specific section, use a heading fragment such as `[systems/payment-router#dependencies]`.

## Persistent Repository Instructions

After successfully consuming a durable repo-local or GitHub-hosted bundle, consider whether persistent repo instructions would help. This is useful when:

- the user says to use the bundle going forward;
- the same bundle is used repeatedly;
- the bundle is a canonical knowledge source for the repository or project;
- future tasks about the repository's domain should start from the bundle.

Do not modify repository-level `AGENTS.md` automatically. Ask first and show the intended behavior. Use `okf_generate_chatgpt_usage` or `generate-chatgpt-usage` to produce the `repo_agents_md_snippet`, then add or update a scoped repository `AGENTS.md` section only after the user explicitly agrees.

The instruction should stay narrow: use the bundle first for relevant domain questions, read `index.md` and `log.md`, prefer OKF MCP/CLI search and context tools, cite concept paths, and report freshness, validation, or coverage gaps.

## Limitations

If the bundle cannot be fetched, say why and suggest one of:

- provide a local path;
- clone the repository into the workspace;
- check GitHub credentials;
- use a public repository URL;
- specify the bundle path inside the repository.