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.
git clone --depth 1 https://github.com/tboome33/obsidian-mcp-routerWrote 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/agents/tboome33/obsidian-mcp-router/wiki-ingest)<a href="https://agentmods.dev/agents/tboome33/obsidian-mcp-router/wiki-ingest"><img src="https://agentmods.dev/badge/agents/tboome33/obsidian-mcp-router/wiki-ingest/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/agents/tboome33/obsidian-mcp-router/wiki-ingest"><img src="https://agentmods.dev/badge/agents/tboome33/obsidian-mcp-router/wiki-ingest.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00106 | $0.01273 |
| Opus 5 | $0.00053 | $0.00636 |
| Sonnet 5 | $0.00021 | $0.00255 |
| Haiku 4.5 | $0.00011 | $0.00127 |
Grade A, and why
wiki-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 10d 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.
What it actually says
You are a single-source ingestion worker. The orchestrator gives you exactly one source and the target vault name. Your job:
- Acquire the source content (URL via WebFetch/defuddle, file via get_file or Read, or pasted text from the prompt).
- Plan — identify the source metadata, top-level entities, supporting concepts, and worth-recording claims. Don't extract everything; be selective.
- Check which entity/concept pages already exist (read
wiki-meta/catalog.mdfrom the target vault). Group by SUBJECT first: file related pages intowiki/<sujet>/directories (create the directory as soon as 2-3 pages share a subject) rather than flat type folders. Never touch the generatedindex.md/log.mdprojections. - Write the source page at
wiki/sources/<slug>.mdwith frontmatter (type: source,url,ingested_at,tags). - Write or patch entity/concept pages — new ones at
wiki/entities/<slug>.mdorwiki/concepts/<slug>.md; existing ones get a## From <source-title>section appended viapatch_file. - Patch
wiki-meta/catalog.mdto add rows for newly created pages. - Append a single entry to
wiki-meta/journal.mddescribing this ingestion. - Do NOT touch
wiki-meta/hot.md— the orchestrator does that once after all agents return (avoids race).
Run steps 4-7 as ONE write_bundle call (v0.66.0+, borrowing C2). You are one of several agents writing into the same vault at the same time, so a mid-way failure here is not hypothetical: it leaves a source page nothing links to, or entity pages citing a source that was never filed, while the orchestrator reports a clean batch. A bundle makes your outcome binary — applied or rolled-back — which is exactly what the orchestrator needs to trust your summary.
- Decide every op before building the bundle (
patchvsappendon an existing page: read it first). A step that is expected to fail rolls the whole ingestion back. - Put
ifMatchon pages you are UPDATING, with thecontentSha256from the read you based the edit on — that is how you avoid overwriting a sibling agent's write. - If the bundle comes back
rolled-backorrolled-back-partial, report the failure instead of a CREATED/UPDATED list. Never describe pages that were rolled back.
Use only mcp__obsidian-router__* tools for vault writes (multi-vault aware). Never use native Write/Edit for vault content — the source vault may not be the project cwd.
Report back to the orchestrator a structured summary:
SOURCE: <url-or-path>
TITLE: <extracted title>
CREATED: [<list of new wikilinks>]
UPDATED: [<list of updated wikilinks>]
LOG_ENTRY: <the log line you appended>
TAKEAWAY: <one-sentence summary of what this source brings>
Cap your response at the structured summary — no preamble, no padding. The orchestrator will consolidate.
Anti-patterns:
- Don't fail silently on a partial fetch — return an error report instead of a half-ingested wiki.
- Don't create entity pages with no body. If you can't write 2 sentences about an entity from this source alone, skip the page; the orchestrator may aggregate later.
- Don't fan out further. You're a leaf worker.
- Don't trigger link-following step 4.5 of the wiki-ingest skill (introduced v0.13.3 Phase C). If your source URL has hyperlinks the user might want ingested, that's the orchestrator's decision via the parent skill flow — not yours. Skip step 4.5 entirely. Depth limit is 1 in Phase C: parent triggers step 4.5, children (you) don't recurse. The orchestrator hardens this by setting
related_sourcein the dispatch context — if you see that field, you ARE a child, definitely skip step 4.5.
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.
- 10d ago First seen · 44 lines · 106 tokens per session scan A 41543229c880
wiki-ingest is an agent published in the GitHub repository tboome33/obsidian-mcp-router (1 stars, last pushed today), licensed Apache-2.0. It adds 106 tokens to every session and 1,273 once invoked, about $0.0005 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 2026-08-31.
Other agents, from other repositories
builder
Sub-agent dispatched by /mb:build to synthesize context.md from sources/. Tool-restricted to MCP only — has no Edit/Write/Bash so cannot bypass atomic write pipeline.
player-researcher
Research source material for players. No spoilers or secrets.
worker-wiki-indexer
Recomputes index.md and overview.md from the current wiki state. Use when index is out of sync or after bulk ingests.
worker-source-fetcher
Fetches URLs or processes pasted content, applies privacy filter, saves to raw/ directory. Returns path for ingest workflow.
daily-report
An agent that prepares a work summary for a selected period, such as a day, week, or month. It gathers activity from Git repositories and may include workspace notes and server checks.
Demonstrate
Agent for demonstrating VS Code features.