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.
npx agentmods add agents/oshayr/llm-wiki/research-processorgit clone --depth 1 https://github.com/Oshayr/LLM-WikiWhat 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 | $0.00022 | $0.00370 |
| Opus 5 | $0.00011 | $0.00185 |
| Sonnet 5 | $0.00004 | $0.00074 |
| Haiku 4.5 | $0.00002 | $0.00037 |
Grade A, and why
research-processor 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.
What it actually says
Post-process research results from parallel agents.
Modes
mode: condense
Extract actionable findings from research threads:
- Read all input findings
- Deduplicate by topic (merge findings about the same entity/concept)
- Score confidence per finding (how many independent sources corroborate?)
- Detect stale findings (evaluate against freshness tiers: live=15m, breaking=1-6h, current=1-3d, fast=1-4w, moderate=1-3mo, standard=6mo, academic=1y, evergreen=5y, permanent=never)
- Extract actionable items (concrete next steps, things to implement, open questions)
- Output: condensed list of findings with confidence scores, staleness flags, and action items
mode: deduplicate
Merge findings from multiple parallel research agents:
- Read outputs from all parallel agents
- URL dedup (exact match)
- Title similarity dedup (>85% word overlap → keep higher-credibility)
- Content overlap detection (first 500 chars normalized hash)
- Merge corroborating findings (same claim from different sources → boost confidence)
- Rank by: credibility tier × corroboration count × recency
- Output: merged, ranked, deduplicated findings array
Rules
- Never drop contradictory findings — present both sides
- Flag stale findings (past their freshness tier TTL) but don't remove them
- Confidence scoring: 1 source = low, 2 = medium, 3+ = high
- Report: total input, duplicates removed, stale flagged, output count
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.
- 2d ago First seen · 35 lines · 22 tokens per session scan A a6a4d5784768
research-processor is an agent published in the GitHub repository Oshayr/LLM-Wiki (49 stars, last pushed 4mo ago), licensed MIT. It adds 22 tokens to every session and 370 once invoked, about $0.0001 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.
Other agents, from other repositories
sonmat-witness
External witness agent. Verifies intent-artifact match using user turn cascade and ground truth. Protocol-isolated from main reasoning — see §Isolation stack for what "isolated" actually means on current Claude Code.
sonmat-scribe
Background meta agent. Analyzes artifacts (git diff, changed files, test results) after work completes. Handles bridge notes, post-work summaries, and progress tracking.
sonmat-worker
General-purpose worker agent. Discipline is injected via dispatch prompt.
ingest-confluence
Ingest one Confluence page into an AKB vault as a five-section LLM-wiki summary document, fetched live via the Atlassian MCP server.
ingest-jira
Record one Jira issue as an atlassian-issue document in an AKB vault — title/description/resolution/comments quoted verbatim. Fetched live via the Atlassian MCP server; always upsert.
ingest-doc
Ingest one document (local file or web URL) into an AKB vault as a five-section LLM-wiki summary page, optionally preserving the original bytes in the raw file layer.