domain-sensemaking

Domain Sensemaking

Use this skill to convert a vague question into an iterative research workflow: frame the problem, generate and rank exploration frontiers, collect evidence, build a concept/claim graph, reframe the question, test convergence, then synthesize a reader-calibrated conclusion.

This skill is platform-neutral. Use whatever capabilities are available: local files, user-provided notes, web search, papers, databases, interviews, code experiments, or only dialogue. If a platform lacks browsing, file writes, or subagents, continue with explicit assumptions and ask for the missing inputs.

Deterministic Helpers

When a filesystem and Python are available, use the bundled scripts/sensemaking_helper.py for fixed-format work instead of recreating tables manually:

python scripts/sensemaking_helper.py init --output path/to/workspace --question "..." --mode Learning
python scripts/sensemaking_helper.py select-frontier path/to/workspace --top 3 --focus "..."
python scripts/sensemaking_helper.py new-round path/to/workspace --focus "..."
python scripts/sensemaking_helper.py new-source path/to/workspace --title "..." --url "..."
python scripts/sensemaking_helper.py score-frontier path/to/frontier.csv --format markdown
python scripts/sensemaking_helper.py lint-workspace path/to/workspace
python scripts/sensemaking_helper.py check-convergence path/to/workspace/convergence.csv --workspace path/to/workspace
python scripts/sensemaking_helper.py record-feedback --workspace path/to/workspace --artifact final-synthesis.md --dimension evidence --verdict negative --tag weak_sources --feedback "..."
python scripts/sensemaking_helper.py summarize-feedback --min-count 3
python scripts/sensemaking_helper.py propose-evolution --min-count 3

All paths above are relative to this skill's directory. Resolve them against the skill's installed location before executing.

• Use init at the start of substantial research to create a problem card, frontier queue, relationship tables, convergence checklist, and synthesis scaffold.
• Use select-frontier after scoring or revising the frontier; it creates the next rounds/round-XX.md from the highest-priority nodes. Manual overrides require an explicit reason.
• Use new-round before each new exploration cycle so the investigation leaves an auditable trail.
• Use new-source whenever you inspect a substantial webpage, paper, report, dataset, interview, or internal document whose contents may need reuse.
• Fill reader-brief.md before writing final-synthesis.md; it controls explanation depth, terminology, structure, and citation expectations for the human-facing artifact.
• Use score-frontier whenever candidate nodes have impact, uncertainty, explorability, and cost scores.
• Treat frontier scores as a scheduling aid, not the final research judgment. Override the ranking when a lower-scoring node is structurally central, blocks many dependencies, or better serves the user's target output.
• Use lint-workspace before final synthesis; fix missing rounds, source notes, empty priorities, or untraceable high-confidence claims before polishing.
• Use check-convergence --workspace before final synthesis to avoid ending with an unchecked narrative or self-declared convergence.
• Use record-feedback when the user critiques a result or when lint/self-review reveals a reusable weakness.
• Use summarize-feedback and propose-evolution after repeated feedback patterns; proposals require human review and tests before changing the skill.
• If scripts cannot run, follow references/templates.md manually and keep the same fields.