thoughtbox:research
Skill64 starsChanged 5 months agoMITView on GitHub
Structured research using Thoughtbox IRCoT (Interleaved Retrieval and Chain-of-Thought). Triggers on: "research", "investigate", "explore", "literature review", "what do we know about".
In this repository102 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- assumptionsSkill
- capture-learningSkill
- Codex-opus-4-6-promptingSkill
- Codex-promptSkill
- mcp-code-modeSkill
- deploy-team-hubSkill
- diagramSkill
- distillSkill
- effect-tsSkill
- escalateSkill
- evalSkill
- experimentSkill
- hddSkill
- hub-auditSkill
- hub-collabSkill
- implementSkill
- knowledgeSkill
- peer-notebook-delivery-guardSkill
- research-taskSkill
- session-reviewSkill
- source-command-hdd-decideSkill
- source-command-hdd-hddSkill
- source-command-hdd-researchSkill
- source-command-hdd-stage-adrSkill
- source-command-hdd-validateSkill
- statusSkill
- supabaseSkill
- synthesizeSkill
- tasteSkill
- teamSkill
- test-suiteSkill
- theseus-protocolSkill
- thoughtbox:debugSkill
- thoughtbox:decisionSkill
- thoughtbox:evolutionSkill
- thoughtbox:knowledge-querySkill
- thoughtbox:onboardSkill
- thoughtbox:refactorSkill
- thoughtbox:session-reviewSkill
- thoughtbox-teamSkill
- ulc-loopSkill
- ulysses-protocolSkill
- workflow-brainstormingSkill
- workflow-ideationSkill
- workflow-reflectionSkill
- workflow-revisionSkill
- workflows-compoundSkill
- workflowSkill
- workflows-planSkill
- workflows-reviewSkill
- workflows-workSkill
- workflow-tournamentSkill
- assumptionsSkill
- capture-learningSkill
- claude-opus-4-6-promptingSkill
- claude-promptSkill
- mcp-code-modeSkill
- deploy-team-hubSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 8 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 21
Create a Thoughtbox session and assess available tools.
- Distinctive: line 43
Use ToolSearch to discover available search tools (Exa, WebSearch, WebFetch, Grep, Read).
- Distinctive: line 45
If no search tools are available, note the limitation and rely on codebase/knowledge graph only.
- Distinctive: line 49
Run this cycle until the question is answered or no productive search avenues remain.
- Distinctive: line 70
Use the best available tool for the current sub-question:
- Distinctive: line 111
Stop the IRCoT loop when ANY of these hold:
- Distinctive: line 113
No more productive search avenues exist
- Distinctive: line 148
Add observations with source attribution to every entity.
- Distinctive: line 180
If the session produced reusable insights, extract them.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Research this topic using Thoughtbox IRCoT: $ARGUMENTS
Protocol References
- Cipher notation for long sessions:
thoughtbox://cipher - IRCoT pattern background:
thoughtbox://interleaved/research
Phase 1: Setup
Create a Thoughtbox session and assess available tools.
// thoughtbox_execute
async () => {
await tb.thought({
thought: "Research question: $ARGUMENTS. Available search tools: [list from ToolSearch]. Strategy: breadth-first scan, then depth on promising leads.",
thoughtType: "context_snapshot",
nextThoughtNeeded: true,
thoughtNumber: 1,
totalThoughts: 30,
sessionTitle: "Research: $ARGUMENTS",
sessionTags: ["research"],
contextData: {
toolsAvailable: [],
constraints: ["primary sources preferred"]
}
});
}Steps:
- Use ToolSearch to discover available search tools (Exa, WebSearch, WebFetch, Grep, Read).
- Record thought 1 as
context_snapshot: state the research question, available tools, constraints. - If no search tools are available, note the limitation and rely on codebase/knowledge graph only.
Phase 2: IRCoT Loop (Think, Search, Integrate, Repeat)
Run this cycle until the question is answered or no productive search avenues remain.
2a: Reason
Record a reasoning thought: what do I need to find out next? What gaps remain?
// thoughtbox_execute
async () => {
await tb.thought({
thought: "Gap: I know X but not Y. Next search: [specific query]. Rationale: [why this fills the gap].",
thoughtType: "reasoning",
nextThoughtNeeded: true,
thoughtNumber: N,
totalThoughts: 30
});
}2b: Search
Use the best available tool for the current sub-question:
- Exa / WebSearch: external knowledge, recent publications, API docs
- Grep / Glob / Read: codebase evidence, local files, prior research
- Knowledge graph:
tb.knowledge.searchEntities()for prior Thoughtbox findings
2c: Integrate
Record findings as a reasoning thought with evidence references. Include source URLs or file paths.
// thoughtbox_execute
async () => {
await tb.thought({
thought: "Finding: [what was learned]. Source: [url/path]. Confidence: [high/medium/low]. Implication: [how this changes understanding].",
thoughtType: "reasoning",
nextThoughtNeeded: true,
thoughtNumber: N,
totalThoughts: 30
});
}2d: Synthesis Checkpoint (every 10-15 thoughts)
Consolidate findings so far. Identify contradictions, gaps, and next priorities.
// thoughtbox_execute
async () => {
await tb.thought({
thought: "Synthesis at thought N: [summary of key findings]. Contradictions: [if any]. Remaining gaps: [list]. Priority for next cycle: [specific question].",
thoughtType: "synthesis",
nextThoughtNeeded: true,
thoughtNumber: N,
totalThoughts: 30
});
}Loop Exit Criteria
Stop the IRCoT loop when ANY of these hold:
- The research question is answered with sufficient evidence
- No more productive search avenues exist
- Thought budget (totalThoughts) is exhausted
Phase 3: Knowledge Persistence
Bridge findings into the Thoughtbox knowledge graph for future retrieval.
For each key finding:
// thoughtbox_execute
async () => {
const entity = await tb.knowledge.createEntity({
name: "Finding: [concise label]",
entityType: "Insight",
observations: [
"[key claim]. Source: [url/path]. Confidence: [level]."
]
});
// Link related findings
await tb.knowledge.createRelation({
from: entity.id,
to: relatedEntityId,
relationType: "BUILDS_ON"
});
}Relation types to use:
BUILDS_ON— finding extends or deepens anotherCONTRADICTS— finding conflicts with anotherRELATES_TO— thematic connection without causal linkSUPPORTS— finding provides evidence for another
Add observations with source attribution to every entity.
Phase 4: Delivery
4a: Conclusion
Record a final thought closing the session.
// thoughtbox_execute
async () => {
await tb.thought({
thought: "Conclusion: [answer to original question]. Key evidence: [top 3 sources]. Confidence: [overall level]. Open questions: [if any].",
thoughtType: "conclusion",
nextThoughtNeeded: false,
thoughtNumber: N,
totalThoughts: N
});
}4b: Export
Present findings to the user as structured markdown:
- Question: the original research question
- Answer: concise answer (2-3 sentences)
- Evidence: numbered list of key findings with sources
- Confidence: overall assessment with reasoning
- Open Questions: what remains unanswered
4c: Extract Learnings
If the session produced reusable insights, extract them.
// thoughtbox_execute
async () => {
const analysis = await tb.session.analyze();
if (analysis.insightDensity > 0.3) {
await tb.session.extractLearnings();
}
}Anti-Patterns
- Searching without reasoning first — always record what you expect to find before searching.
- Recording every intermediate step — only record findings that change understanding.
- Skipping synthesis checkpoints — drift compounds; synthesize every 10-15 thoughts.
- Orphan knowledge — every entity needs at least one relation. Isolated nodes are unfindable.
- Premature conclusion — do not conclude until you have checked for contradictory evidence.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

