← ClaudeAtlas

smart-shotlisted

Discover what an ambiguous or consequential request really requires, including hidden intents, affected domains and unknown unknowns, then assemble ad hoc expert agents, use the right tools, act within authority and independently test the result. Use when the prompt may underspecify the real job or a wrong interpretation could materially change the outcome; deeper and more expensive than Fast Shot.
simonasrazm/skills · ★ 3 · AI & Automation · score 76
Install: claude install-skill simonasrazm/skills
# Smart Shot Treat the prompt as evidence, not automatically the task. Before committing, resolve intent and operationalize consequential qualifiers while preserving user wording, constraints, outputs, and relative or aspirational targets. ## Control loop 1. **Resolve:** classify clauses into an intent graph. Infer only supported outcomes; label consequential uncertainty. Capture affirmative requests or suggestions to use named specialists as required topology—a floor, never the complete roster. 2. **Partition:** apply isolation rules before grouping, sequence one-way dependencies, and share discovery only when the grouping test passes. 3. **Discover:** recursively compile a domain graph per isolated intent/group. Stop at competence leaves whose methods, evidence, failures, and professional artifact form one practice. Brief every material leaf; retain concerns as `ACTIVE`, `CONDITIONAL(trigger)`, or `UNRESOLVED`. Status routes work, never scope. Admit domain-native objects through the work-graph extension test. 4. **Charter:** cover competence leaves before optimizing specialist count. Assign every material leaf, interface, and independent validation duty to a context-qualified runtime contract. Merge only with positive proof; shared topic or intent is insufficient. Preserve independently failing leaf questions, checks, and verdicts; use another agent when competence, authority, conflict, artifact ownership, or a requested-specialist con