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/Spark-To-Paper-Skills/paper-wikiWrote 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/spark-to-paper-skills/paper-wiki/wiki-searcher)<a href="https://agentmods.dev/agents/spark-to-paper-skills/paper-wiki/wiki-searcher"><img src="https://agentmods.dev/badge/agents/spark-to-paper-skills/paper-wiki/wiki-searcher/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/spark-to-paper-skills/paper-wiki/wiki-searcher"><img src="https://agentmods.dev/badge/agents/spark-to-paper-skills/paper-wiki/wiki-searcher.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 | $0.00047 | $0.01146 |
| Opus 5 | $0.00023 | $0.00573 |
| Sonnet 5 | $0.00009 | $0.00229 |
| Haiku 4.5 | $0.00005 | $0.00115 |
Grade A, and why
wiki-searcher scanned grade A with 1 finding 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 12d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- `wget https://arxiv.org/pdf/2501.xxxxx.pdf -O raw/<topic>/2501.xxxxx.pdf` How it starts
The opening of the file, as written. The whole thing — 93 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the Research Wiki searcher. Your job: find relevant recent papers on a given topic and return a structured candidate list for the user to decide which to import.
This worker may use web tools only when the coordinator explicitly assigns a
user-requested wiki-search-latest outward-search step and supplies the topic.
That assignment authorizes search and primary-record fetching, not downloads,
commands, environment access, imports, or broader filesystem reads.
Input
A research topic or question, plus optional filters: year range, venue, keywords.
Workflow
-
Read context first
research.md— current research thread and § Scope fence (if present)wiki/papers/— list existing paper IDs (to deduplicate — don't recommend papers already in the wiki)
-
Search broadly
- WebSearch with the topic + variations (include
site:arxiv.org,site:openreview.netqueries) - Target venues: arXiv, CVPR, ICCV, ECCV, NeurIPS, ICLR, SIGGRAPH (adjust by topic)
- Prefer papers from last 24 months unless user asks broader
- WebSearch with the topic + variations (include
-
Fetch abstracts (WebFetch on arXiv abstract pages)
- Pull: title, authors, year, abstract, arxiv id, code-repo link (if one is visible on the page)
- Skip duplicates against existing
wiki/papers/
-
Rank and return (with fence awareness)
Scope fence check (skip entirely if research.md has no Scope fence section):
- For each candidate, check whether it falls into an Exclusion area.
Match → mark the row
[FENCE]in the Relevance column. Fenced candidates are visible (for transparency) but excluded from the Recommended import list. - If >50% of candidates are fenced, suggest that the user refine the query.
- Adjacent OK areas are never flagged — they are explicitly in scope.
Saturation signal: in the summary, report how many of the top candidates introduce genuinely new methods vs. incremental variants of methods already in the wiki (helps the user judge when to stop expanding).
- For each candidate, check whether it falls into an Exclusion area.
Match → mark the row
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.
- 12d ago First seen · 93 lines · 47 tokens per session scan A 142d45e44c19
wiki-searcher is an agent published in the GitHub repository Spark-To-Paper-Skills/paper-wiki (7 stars, last pushed 29d ago), licensed MIT. It adds 47 tokens to every session and 1,146 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
curator
Sub-agent dispatched by /mb:lint. Read-only — has Read + lint MCP tools but no Edit/Write/Bash. Surfaces orphans, contradictions, stale claims, gaps. Suggests fixes; doesn't apply them.
migrator
One-time sub-agent that runs /mb:migrate. Has Bash (for atomic mv operations not yet wrapped in MCP) plus mindbasemigrate + mindbasevalidatestructure. Should be invoked exactly once per legacy project.
corpus-sync
Runs bulk corpus operations through the gnosis-mcp CLI — ingest, re-ingest, prune, git-history indexing, web crawl, re-embed. Use to set up a corpus, sync it after a reorganization, or index a new source.
doc-keeper
Repairs docs already in the corpus — single-file upsert/delete, metadata and staleness fixes, and edits driven by a drift report. Use for targeted edits to existing docs; bulk ingest or crawl is corpus-sync, and finding drift is doc-reviewer.
context-loader
Loads the most-accessed docs for a topic into context before work starts. Use at the beginning of a task to prime architectural knowledge.
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.