A portable research-intelligence workflow for disciplined discovery and cross-domain structural abstraction — for problems where ordinary summarization or brainstorming is insufficient.
Two complementary protocols packaged as one reusable Skill, kept operationally distinct. A task that only requires evidence acquisition stops after Discovery. A task that begins with an existing evidence base can invoke AoA directly. A full workflow runs both in sequence.
Governs the path from reality, observations, and experiments to structured evidence — improving evidence acquisition and hypothesis discrimination.
Prefer observations and experiments that distinguish among competing explanations over observations that merely confirm the current favorite.
Governs the path from structured evidence to tested invariants, cross-domain transfer, and architecture-ready knowledge — without confusing analogy with equivalence.
Candidate abstractions must compete, survive structural discrimination, preserve provenance and uncertainty, and state transfer limits explicitly.
AoA assigns the strongest supported level — and no higher. Superficial resemblance is never presented as functional, dynamic, governed, or formal equivalence.
Keeping the components separate reduces a common failure mode in AI-assisted research: abstracting too early from observations that have not earned sufficient evidentiary status.
What evidence should we acquire next, and what does that evidence justify?
What reusable structure survives across the evidence, and how far can that structure legitimately transfer?
Prioritizes experiments that separate competing explanations rather than accumulating support for one favored theory.
Unknowns, confounds, assumptions, and surviving alternatives stay visible instead of collapsing into a single narrative.
Rejected hypotheses and failed experiments remain part of the record. Conclusions revise without rewriting history.
Distinguishes observed material, derived structure, candidate abstractions, supported invariants, speculation, and rejected mappings.
The equivalence ladder keeps cross-domain mappings honest about how much structure actually carries over.
Discovery alone, AoA alone, or both in sequence — the Skill does not run the full pipeline when the task needs one component.
Structure observations, generate competing explanations, and find the next experiment with the highest discriminatory value.
Distinguish plausible mechanisms with controlled tests rather than arguing from narrative coherence.
Test whether structures in one domain transfer to another at the relational, functional, dynamic, governance, or formal level.
Map source and target roles, preserved relations, known mismatches, transfer assumptions, and falsifiers.
Identify structural invariants while keeping observed and derived material separate.
Discovery produces a structured evidence corpus; AoA then tests which structures deserve compression and transfer.
Especially useful for competing hypotheses, weak or incomplete evidence, cross-domain analogy, causal uncertainty, or the temptation to compress a complex system too early.
Both components support compact, standard, and deep output modes. Depth is driven by the research problem, not by a desire for longer prose.
Distributed as an open SKILL.md-based package. The guide distinguishes native Agent Skill installation from compatibility approaches on platforms that use a different customization mechanism.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Platform-specific steps in the installation guide.
Open, modular, and versioned. Fork it, test it, and benchmark it against your own research.