Install
$ agentstack add skill-v0lka-skills-research-hypothesis ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Hypothesis Management
Create, update, and manage hypothesis cards and the hypothesis graph for an research project.
Context
The hypothesis is the atomic unit of work in the methodology. Hypotheses are organized as a directed acyclic graph (DAG) where edges represent the relationship "result of X led to formulation of Y." The graph grows as the research progresses — only initial hypotheses are known at the start.
Each hypothesis has a card (a dedicated .md file) and an entry in the graph page (hypotheses/graph.md).
Hypothesis Card Fields
Every hypothesis card contains these fields:
| Field | Description | |---|---| | Identifier | Unique ID within the research, e.g., H-001 | | Short title | Concise label for graph nodes and catalog entries | | Statement | What is being tested, phrased as a falsifiable assertion | | Verification criterion | Specific condition for confirmation (working PoC, metric threshold, counterexample) | | Timebox | Maximum time allocated for verification (ISO 8601 duration or calendar dates) | | Parents | Links to hypothesis cards whose results led to this one (empty for root hypotheses) | | Status | open / in-progress / confirmed / refuted / cancelled | | Created | Date the hypothesis was formulated (ISO 8601) | | Completed | Date the experiment concluded (filled upon completion) | | Decision | The iteration decision applied after the result: continue / pivot / kill / fork (filled by research-decision) | | Experiment notes | Free-form observations, intermediate findings, and methodology notes recorded during the experiment | | Result | Filled upon completion: findings, link to prototype or proof, list of artifacts |
Operations
Create a New Hypothesis
Step 1 — Gather information
Collect from the researcher:
- Statement — a clear, falsifiable assertion. Good: "Static analysis of
webpack bundles can extract >= 90% of API endpoints." Bad: "Look at webpack bundles."
- Verification criterion — what constitutes confirmation. Must be
specific and measurable.
- Timebox — maximum time for this hypothesis. Push back if the timebox
seems too generous (encourages scope creep) or too tight (guarantees inconclusive results).
- Parents — which existing hypothesis results motivated this one. For
root hypotheses (formulated during Exploration), parents are empty.
If the statement is vague, help the researcher sharpen it. A hypothesis must be an assertion that can be confirmed or refuted, not a topic or direction.
Step 2 — Assign identifier
Read hypotheses/graph.md to find the highest existing H-NNN. Assign H-(N+1), zero-padded to three digits.
Step 3 — Create the card file
Create hypotheses/H-NNN.md using the template at [assets/hypothesis-card-template.md](assets/hypothesis-card-template.md). Fill in all collected fields. Set status to open.
Step 4 — Update the graph
Update hypotheses/graph.md in two places:
- Mermaid diagram: Add a new node with the appropriate CSS class
(open for new hypotheses). If the hypothesis has parents, add edges from each parent node to the new node.
- Catalog table: Add a row with ID (linked to the card file), statement
summary, status, decision (empty for new), and parent IDs.
Update Hypothesis Status
When a hypothesis transitions between statuses, update:
- The card file (
H-NNN.md): change the Status field. - The graph diagram: change the CSS class of the node.
- The catalog table: update the Status column.
Valid transitions:
open → in-progress → confirmed
→ refuted
→ cancelled
open → cancelled
A hypothesis cannot transition backwards. If rework is needed, create a new hypothesis with the original as parent.
Record Results
When an experiment completes (see research-experiment skill), update the hypothesis card:
- Set Status to
confirmed,refuted, or leave asin-progressif
the result is inconclusive.
- Fill the Result section with:
- What was discovered.
- Link to the prototype, proof, or experiment artifacts.
- If inconclusive: why, and what would be needed for a definitive result.
- Update the graph diagram and catalog table accordingly.
Fork Hypotheses
When a fork decision is made (see research-decision skill), create multiple child hypotheses from the same parent. Each competing approach becomes its own hypothesis with its own timebox and verification criteria. Document in each card that it is part of a fork and which hypotheses are its siblings.
Mermaid Diagram Conventions
Use this node styling:
classDef confirmed fill:#4CAF50,color:#fff
classDef refuted fill:#F44336,color:#fff
classDef in_progress fill:#FF9800,color:#fff
classDef open fill:#2196F3,color:#fff
classDef cancelled fill:#9E9E9E,color:#fff
Node IDs use the pattern H001, H002, etc. (no hyphen, for Mermaid compatibility). Display labels include the hyphen: H-001: Short title.
Edges represent "result led to formulation":
H001 --> H003
H002 --> H003
Important Notes
- One hypothesis at a time is the default. Fork (parallel hypotheses)
is the exception, used only when competing approaches are discovered and resources allow parallel investigation.
- Timebox is a hard ceiling: if the timebox expires without a result,
that is itself a result (hypothesis not confirmed in allocated time) and a signal for kill or pivot.
- Keep hypothesis statements atomic. If a statement contains "and," consider
splitting into two hypotheses.
- The graph is the primary navigation tool for the research. Keep it accurate
and up to date at all times.
Relation to Other Skills
- Invoked by
research-initto create the first hypothesis. - Works closely with
research-experiment(records results) and
research-decision (determines next hypotheses after a result).
research-statusreads the graph to generate overviews.research-synthesisuses the graph to produce the final report.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: v0lka
- Source: v0lka/skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.