Assumption Excavation

Find what the artifact takes for granted before those assumptions become implementation bugs.

Use When

• reviewing PRDs, specs, plans, or issue slices before work starts;
• reviewing AGENTS.md, SKILL.md, SET bundles, proof-loop contracts, or workflow docs;
• a proposal sounds confident but depends on unclear prerequisites;
• you need a lightweight pre-implementation review lens.

Workflow

1. Identify the artifact and its intended consumer.
2. Read for implicit assumptions, not prose quality.
3. Classify assumptions:
   • environment: tools, services, files, auth, OS, network;
   • dependency: upstream state, input shape, prior checks;
   • behavioral: what humans, agents, or systems are expected to do;
   • temporal: what must remain stable over time;
   • success: what the artifact counts as done or safe.
4. Rank the assumptions by fragility:
   • high: likely to break or expensive if wrong;
   • medium: plausible risk with manageable fallout;
   • low: visible but not blocking.
5. Mark partially surfaced assumptions as PARTIAL.
6. Recommend what must be clarified, tested, or accepted before implementation.

Output Shape

Use:

• verdict: EXAMINED or UNEXAMINED;
• top assumptions, ranked;
• fragility score or severity per item;
• missing evidence;
• smallest clarification or verification step.

Guardrails

• Do not rewrite the artifact unless asked.
• Do not treat assumptions as errors by default; some are deliberate.
• Do not block on low-risk assumptions.
• State uncertainty as text-based inference, not mind reading.

Provenance

Adapted from the assumption-excavator pattern in aself101/agents-and-pipelines, compressed for ABVX skill use and broader repo-contract review.