An Explainer · v0.1
Discourse Has Weather.
An ecological reading of the InfraNodus v2 skill — why thinking is a spiral, not a pipeline, and how to drive it on purpose.
01 · The Premise
Stop running pipelines. Start checking the weather.
The old InfraNodus skill was a pipeline: analyze, find gaps, generate questions, done. Same sequence every time, regardless of what the text actually looked like. The v2 skill replaces that with an ecological loop borrowed from Panarchy theory — the cycle of growth, saturation, release, and reorganization that ecosystems pass through.
Every analysis follows a recursive spiral. You sense the discourse state first, pick tools based on what you find, then re-sense to see what shifted. The agent's job is not to execute a sequence — it's to drive the spiral toward productive states.
This explainer walks through the five elements that make the v2 skill an actual decision protocol instead of a checklist: the spiral itself, the six discourse states it diagnoses, the transition map between them, the tools that drive each transition, and the failure mode (stagnation) that the whole architecture exists to prevent.
02 · The Spiral
Five steps. No fixed length. The loop is the point.
Sense, then diagnose, then intervene, then bridge, then re-sense. Each cycle generates knowledge, which generates new questions, which require new tool runs. You don't finish — you stop when the discourse is in a productive state.
What each step does
Sense — run generate_knowledge_graph to get the structural readout: modularity, cluster count, top-node betweenness, diversity score.
Diagnose — read the full output, not just numbers. The metrics can say "diversified" while the too_focused_on_top_clusters flag says otherwise. Trust the flags.
Intervene — look up the diagnosis in the Adaptive Response Table (Section 03 below) and run the prescribed primary + secondary tools.
Bridge — develop_conceptual_bridges runs every cycle. v2 promotes it from optional to core practice. It connects the local discourse to broader context.
Re-sense — re-run the graph after intervention. Did the state shift? If yes, loop with the new diagnosis. If no for two cycles in a row, you've hit stagnation — switch to escape mechanisms.
03 · The Six Discourse States
The diagnosis tells you which protocol to run.
Each state has a metric signature (what the modularity and cluster numbers look like) and a prescription (which tools run as primary and secondary intervention). The first job is matching what you see to one of these.
Biased
develop_latent_topics — surface what's been suppressed.generate_content_gaps at higher gapDepth.modifyAnalyzedText to extractEntitiesOnly to break word dominance.Focused
generate_content_gaps + develop_conceptual_bridgesgenerate_research_questions with useSeveralGaps: truegapDepth to find deeper structural holes.Diversified
generate_research_questions per cluster — deepen each.overlap_between_texts to find convergence threads.Dispersed
overlap_between_texts — find common threads.generate_contextual_hint for synthesis.Transitional
develop_conceptual_bridges — lock in the reorganization.generate_research_ideas with extended ideation mode.Stagnant — failure mode
modifyAnalyzedText.modelToUse.gapDepth or remove the top three nodes.04 · State Map
The transitions are the point.
Diagnoses are not destinations. Each state has a default trajectory toward another state — and v2 names the mechanism. Click a state to see where it goes and what gets it there.
The state map
Each node is a discourse state. Each arrow is the transition v2 prescribes — labeled with the mechanism that drives it. Click any node to highlight what flows into and out of it.
05 · The Transition Table
How states change — and how to drive the change on purpose.
The map above is the visual. The table below is the protocol. Each row is a transition: the mechanism that produces it and the tool sequence that executes it.
| From → To | Mechanism | Tools |
|---|---|---|
| Biased → Focused | Introduce peripheral concepts. Remove dominant-node influence. | develop_latent_topics → add latent content → re-analyze |
| Focused → Diversified | Bridge structural gaps with genuinely new content (not just reframing). | generate_content_gaps → generate_research_ideas → add bridging content → re-analyze |
| Diversified → Productive | Deepen each cluster with specifics. Connect the most promising pairs. | generate_research_questions per cluster → develop each → re-analyze |
| Dispersed → Convergent | Find overlap. Create synthesis notes. Build shared vocabulary. | overlap_between_texts → develop_conceptual_bridges → synthesize |
| Any → Stagnant | Warning sign. Re-analysis produces the same result twice in a row. | Change modifyAnalyzedText · increase gapDepth · switch modelToUse · remove top 3 nodes |
What a real transition looks like in metrics
From the live demo in Section 3 of the source walkthrough — adding bridging content to a "focused" graph and re-sensing:
| Metric | Before | After bridging | Direction |
|---|---|---|---|
| Modularity | 0.63 | 0.54 | ↓ clusters merged |
| Cluster count | 8 | 6 | ↓ consolidated |
| Top cluster % | 29% | 27% | ↓ less dominant |
| Top node BC | 0.34 (discourse) | 0.33 (discourse) | → stable |
| too_focused_on_top_clusters | true | false | ✓ flag flipped |
| Top node influence ratio | 0.24 | 0.19 | ↓ more distributed |
| Diversity score | focused | focused | → healthier, not transitioned |
One round moved the needle but didn't complete the transition. Diversity score is still "focused." This is the iterative truth of the spiral: transitions take more than one cycle. You intervene, re-sense, and if the state hasn't fully moved, you intervene again.
06 · Tool Inventory
Twenty-four tools. Six categories. One mental model: cost gating.
The toolbox you draw from when the Adaptive Response Table tells you what to run. The big v2 promotion: develop_conceptual_bridges is now core practice — run it every workflow, not just when you remember.
01 Analysis · text → insight
The sensing and diagnosing tools. Take text in, get structure back.
- generate_knowledge_graphFull network analysis. Always first — the Sense step.
- generate_topical_clustersJust the clusters. Cheap gate check.
- generate_content_gapsStructural holes between clusters. Primary intervention for focused/biased.
- generate_contextual_hintTopical overview for RAG. Lightweight prompt augmentation.
- develop_text_toolCombined: gaps + questions + latent topics. The expensive all-in-one.
- develop_conceptual_bridgesConnect text to broader discourse. EVERY workflow.
- develop_latent_topicsUnderdeveloped threads. Primary intervention for biased state.
02 Creation & Persistence
For building things that persist across sessions.
- create_knowledge_graphSave a named graph to your InfraNodus account.
- memory_add_relationsSave entity relations to persistent memory.
- memory_get_relationsRecall saved relations by entity or context.
03 Comparison · text vs text
When you have two or more things to compare.
- overlap_between_textsWhat do they share? Primary tool for dispersed state.
- difference_between_textsWhat's in B but missing from A?
04 Retrieval & Search
Working with saved graphs and existing knowledge.
- analyze_existing_graph_by_namePull and analyze a saved graph.
- retrieve_from_knowledge_baseGraphRAG — query with natural language.
- searchFind concepts across all your graphs.
- fetchGet a specific search result by ID.
- generate_responses_from_graphUse a graph as an expert knowledge base.
- research_questions_from_graphGenerate questions from a saved graph's gaps.
05 SEO & Search Intelligence
Market and content opportunity analysis.
- analyze_google_search_resultsWhat people find — supply.
- analyze_related_search_queriesWhat people search for — demand.
- search_queries_vs_search_resultsThe gap between demand and supply.
- generate_seo_reportCompare your content against the landscape.
06 Research & Ideation
For generating novel directions.
- generate_research_questionsQuestions from structural gaps.
- generate_research_ideasIdeas from the same gaps.
07 · The Adaptive Workflow
Deep text analysis, end to end.
The primary workflow — what you run when someone hands you text and says "analyze this." The branching logic in steps 3 and 4 is what makes v2 adaptive. Old workflow ran the same four tools every time. v2 picks the right tools for the actual state.
Sense
Run the graph. Read modularity, cluster count, top-node betweenness, diversity score, and the boolean flags.
generate_knowledge_graphDiagnose
Match the readout to one of the six states. The flags can override the numbers — when too_focused_on_top_clusters is true, you're focused even if modularity reads diversified.
Branch — choose the prescription
Look up the state in the Adaptive Response Table. Run the primary intervention. Run the secondary if you have budget.
generate_content_gapsdevelop_latent_topicsoverlap_between_textsgenerate_research_questionsAlways bridge
Run conceptual bridges every workflow, regardless of state. v2 promoted this from optional to core. It's how the local discourse gets connected to broader context.
develop_conceptual_bridgesRe-sense
Re-run the graph on the extended content. Did the state shift? If yes, loop with the new diagnosis. If no for two cycles in a row, you've hit stagnation — switch to escape mechanisms (Section 08).
generate_knowledge_graph08 · Stagnation & Escape
The whole architecture exists to prevent fixed points.
Stagnation is the failure mode. When re-analysis produces the same clusters and gaps, the discourse has hit a fixed point — the spiral isn't moving anything. Sameness = symmetry = stuck. Each escape mechanism works by injecting asymmetry.
A · Scale Shift
Change the lens.
Switch modifyAnalyzedText mode — for example, from none to extractEntitiesOnly.
Like zooming out on a map. Different patterns become visible at different scales.
B · Perspective Shift
Change the model.
Switch modelToUse — Claude → GPT-5 → Gemini.
Different models surface different latent structure. The text didn't change; the reading did.
C · Structural Shift
Change the depth.
Increase gapDepth. Enable useSeveralGaps. Mentally remove the top three nodes.
Forces the analysis past the obvious.
Why this works
Stagnation is symmetry. The system has reached an equilibrium where the same inputs produce the same outputs. Each escape mechanism breaks a different symmetry: scale (the resolution), perspective (the reader), and structure (the priority order). Inject any one of those and the system has to re-organize.
09 · Open Threads
What's left to build.
Two things from the source walkthrough that should be picked up next. Both are about extending the spiral from single-document analysis to a continuously-running practice.
create_knowledge_graph + memory_add_relations so the analysis lives somewhere durable and the result is shareable.generate_content_gaps is a release intervention. develop_conceptual_bridges is a reorganization intervention. If the spiral is real (and the metrics say it is), then any piece of intellectual work — not just InfraNodus runs — is moving through these phases. The skill names them. What other practices could be re-described in this vocabulary?