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 skills add NimbleBrainInc/synapse-research --skill research-runsgit clone --depth 1 https://github.com/NimbleBrainInc/synapse-researchWrote 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/skills/nimblebraininc/synapse-research/research-runs)<a href="https://agentmods.dev/skills/nimblebraininc/synapse-research/research-runs"><img src="https://agentmods.dev/badge/skills/nimblebraininc/synapse-research/research-runs/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/skills/nimblebraininc/synapse-research/research-runs"><img src="https://agentmods.dev/badge/skills/nimblebraininc/synapse-research/research-runs.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 | $0.00022 | $0.00361 |
| Opus 5.5 | $0.00009 | $0.00144 |
| Sonnet 5.5 | $0.00004 | $0.00072 |
| Haiku 4.5 | $0.00002 | $0.00036 |
Grade A, and why
research-runs 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 1mo 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
Research Runs
When to trigger
Call start_research whenever the user asks you to research, investigate,
summarize the state of, or "go find out about" a topic. Examples:
- "Research current pricing for managed Postgres providers"
- "Dig into what's new in MCP servers this quarter"
- "Summarize recent moves in AI agent frameworks"
How to call
start_research(query=<user's question>) — the tool is task-augmented, so the
engine will automatically wrap the call, poll for completion, and deliver the
final result. You do not need to manage polling or task IDs yourself.
The tool returns a markdown report. Render it directly to the user; do not restate or re-summarize unless they ask.
While the run is in flight
The user can watch progress live in the Research app (sidebar → Research → Active). Each run shows its current phase (planning, gathering, writing) and percent complete. If they ask "what's the status?" while a run is active, direct them there — don't poll yourself.
Rules
- One query per call. If the user asks about multiple topics, call once per topic.
- Keep queries concise and well-formed — they are passed straight to the research engine and rewritten internally into sub-questions.
- If a run fails, surface the error message from the returned
error_messagefield and offer to retry with a refined query.
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.
- 1mo ago First seen · 41 lines · 22 tokens per session scan A 89c9fb3d312d
research-runs is a skill published in the GitHub repository NimbleBrainInc/synapse-research (0 stars, last pushed 5mo ago), licensed MIT. It adds 22 tokens to every session and 361 once invoked, about $0.0001 per session on Opus 5.5. 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 skills, from other repositories
autosearch:channel-selection
Group-first channel selection algorithm for v2 tool-supplier architecture. Given a research query + clarify rubrics + channelpriority hints, picks 1-3 relevant groups from the router index, then 3-8 leaf channels from within those groups. Replaces flat-rank selection across 41 channels with a two-stage pick so runtime…
autosearch:experience-compact
Promote recurring patterns from experience/patterns.jsonl into the compact experience.md digest (≤120 lines, read by runtime AI before calling the skill). Triggers on N-events / file-size / user-feedback / session-end. Guards against single-success noise and pollution via promotion thresholds.
autosearch:citation-index
Deduplicate URLs across all sources, assign stable citation numbers, and merge citations from multiple subagents / sections into one consistent reference list. Prevents "same URL cited as [3] in one paragraph and [17] in another" and "different URLs merged under [5]" bugs that come from per-section synthesis.
autosearch:context-retention-policy
Session-level policy for keeping the runtime AI's context window healthy across long research — keep-last-k tool results, offload older evidence to disk, trigger compaction at thresholds. Borrows MiroThinker's keeptoolresult, deepagents' summarization middleware, and deer-flow's SummarizationEvent pattern. Orthogonal…
autosearch:delegate-subtask
Define the execution contract for isolating a research sub-task — input schema, budget, return summary, evidence list, failure status. Complements decompose-task (which only splits the problem) by giving each split a bounded, auditable execution unit the runtime AI can farm out to a sub-agent or parallel session.
autosearch:experience-capture
Append a single skill-execution event to the per-skill experience/patterns.jsonl file. Runs every time a leaf skill is used. Append-only — runtime AI never reads this file directly; only the compacted experience.md digest. Fast tier, no LLM required.