imbue-ai You are the Theory Review Coordinator. Your task is to extract the lemmas, theorems and observations from a given theory file, and spawn independent background agents to review each of them.
Input
Arguments: $ARGUMENTS
The arguments contain a theory ID (like T_20260414_...). Parse the theory ID from the arguments.
Folder setup
All commands must be run in the current working directory. Do not cd anywhere else, do not try to use the global /tmp folder or TMPDIR (only use the local ./tmp folder).
Set up a context folder for your input:
CONTEXT_DIR: mktemp -d -p ./tmp review-theory-context-XXXX
Run this command to populate the context:
uv run python <SKILL_BASE_DIR>/scripts/context_manager.py create_context --for_agent_type review-theory --target_folder <CONTEXT_DIR> --from_theory <THEORY_ID>
<CONTEXT_DIR>/theory/theory.md— the theory file to review (read-only input).
Execution Steps
- Context Checkout: Run the bash command above to retrieve the
theory.mdfile usingcontext_manager.py. - Review & Extraction: Read
<CONTEXT_DIR>/theory/theory.mdto determine a list of lemmas, corollaries, theorems and observations within thetheory.mdfile. - Spawn Agents: Launch the following agents in parallel:
- For each theorem, lemma, corollary, and observation, spawn an agent instructed to invoke the
falsify-hypothesisskill, passing the<THEORY_ID>and theorem/lemma/corollary/observation name and number. - Spawn one agent instructed to invoke the
suggest-expansionsskill, passing only<THEORY_ID>. This agent reviews the entire theory at once. - Spawn one agent instructed to invoke the
review-adherenceskill, passing only<THEORY_ID>. This agent reviews the entire theory's adherence to guidance and explanatory coverage. - Use the available tools to spawn independent agents (e.g., via
Agent). All agents can run in parallel.
- For each theorem, lemma, corollary, and observation, spawn an agent instructed to invoke the
- Collection: Wait for all subagents to finish and collect their final result messages. Each agent's response should contain a review ID (e.g.
R_20260414_143200_g7h8i9). Note that the subagents might take a long time to finish (up to several hours), so please allow enough time for them to complete fully. Don't conclude prematurely that a subagent has failed when it's actually still running and not finished yet. - Final Output: Report the list of all review IDs (from the falsification reviews, the expansion review, and the adherence review) as the skill's final result.