/propose-subgoal-decomposition-plans
Propose multiple subgoal decomposition plans for the current theorem using the information already gathered. Use when enough information has been collected from examples, counterexamples, search results, and previous failures to break the problem into several materially
$ npx -y skills add frenzymath/Danus --skill propose-subgoal-decomposition-plans --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
- Slash command
/propose-subgoal-decomposition-plans
Context preview
The summary Claude sees to decide when to auto-load this skill.
Propose multiple subgoal decomposition plans for the current theorem using the information already gathered. Use when enough information has been collected from examples, counterexamples, search results, and previous failures to break the problem into several materially
SKILL.md
propose-subgoal-decomposition-plans.SKILL.mdname: propose-subgoal-decomposition-plans
description: Propose multiple subgoal decomposition plans for the current theorem using the information already gathered. Use when enough information has been collected from examples, counterexamples, search results, and previous failures to break the problem into several materially different plans.
Propose Subgoal Decomposition Plans
Use this skill when the agent has enough context to propose several viable decomposition plans.
Input Contract
Read:
- the current target theorem or branch goal
- relevant `immediate_conclusions`, `toy_examples`, and `counterexamples`
- relevant `failed_paths` and `branch_states`
- recent search results and useful references from `events`
Procedure
1. Gather the current information that materially constrains the problem: useful examples, failed claims, known obstructions, and relevant search results. 2. Propose materially different decomposition plans. 3. For each plan, state:
- the main idea of the plan
- the ordered subgoals
- why this plan is plausible given the current information
- which earlier failures or counterexamples it tries to avoid
4. Hand each plan to `$direct-proving` for a quick screening pass.
Output Contract
Publish one plan per decomposition to global memory with `gm_add` (kind `plan`, a judgment — `verifiable=false`): `claim` = the plan's goal + summary, `evidence` = its motivation, plus these fields:
{
"plan_id": "...",
"record_type": "decomposition_plan",
"goal": "...",
"plan_summary": "...",
"subgoals": ["..."],
"motivation": ["..."],
"uses_information_from": {
"examples": ["..."],
"counterexamples": ["..."],
"key_failures": ["..."],
"search_results": ["..."]
},
"status": "proposed|screening|screened|selected|failed|solved",
"branch_id": "optional"
}Also note the new plan set in your local memory (`events`).
Tools
- `gm_add` (publish the plan findings)
- `gm_search` (recall the examples/counterexamples/dead-ends the plans build on)
- `search_arxiv_theorems`
Failure Logging
If the agent cannot yet propose meaningful decomposition plans, append an `events` record with:
- `event_type="decomposition_plans_not_ready"`
- the missing information
- the blockers that prevent proposing plans
Read more
name: propose-subgoal-decomposition-plans description: Propose multiple subgoal decomposition plans for the current theorem using the information already gathered. Use when enough information has been collected from examples, counterexamples, search results, and previous failures to break the problem into several materially different plans.
Propose Subgoal Decomposition Plans
Use this skill when the agent has enough context to propose several viable decomposition plans.
Input Contract
Read:
- the current target theorem or branch goal
- relevant `immediate_conclusions`, `toy_examples`, and `counterexamples`
- relevant `failed_paths` and `branch_states`
- recent search results and useful references from `events`
Procedure
1. Gather the current information that materially constrains the problem: useful examples, failed claims, known obstructions, and relevant search results. 2. Propose materially different decomposition plans. 3. For each plan, state:
- the main idea of the plan
- the ordered subgoals
- why this plan is plausible given the current information
- which earlier failures or counterexamples it tries to avoid
4. Hand each plan to `$direct-proving` for a quick screening pass.
Output Contract
Publish one plan per decomposition to global memory with `gm_add` (kind `plan`, a judgment — `verifiable=false`): `claim` = the plan's goal + summary, `evidence` = its motivation, plus these fields:
{
"plan_id": "...",
"record_type": "decomposition_plan",
"goal": "...",
"plan_summary": "...",
"subgoals": ["..."],
"motivation": ["..."],
"uses_information_from": {
"examples": ["..."],
"counterexamples": ["..."],
"key_failures": ["..."],
"search_results": ["..."]
},
"status": "proposed|screening|screened|selected|failed|solved",
"branch_id": "optional"
}Also note the new plan set in your local memory (`events`).
Tools
- `gm_add` (publish the plan findings)
- `gm_search` (recall the examples/counterexamples/dead-ends the plans build on)
- `search_arxiv_theorems`
Failure Logging
If the agent cannot yet propose meaningful decomposition plans, append an `events` record with:
- `event_type="decomposition_plans_not_ready"`
- the missing information
- the blockers that prevent proposing plans
Danus orchestrates mathematical reasoning agents with fact-graph memory. A main agent (Claude Code) steers a swarm of autonomous codex workers that prove; a cold-start verifier is the sole authority on correctness: a result becomes real only once it passes.
Other skills on danus.
- /consult
Consult a strong reasoning model for strategy — feed it the current elaboration, take its reply as the next master_guidance, and dispatch workers from it. This is the main agent's high-intelligence step (workers prove; the consult decomposes and steers). Runs over the gpt_pro
Open skill - /elaboration
Write a high-signal-to-noise mathematical progress synthesis of a project from the shared stores (global memory + fact graph), following a fixed verdict→routes→interfaces→dangers→bridges template and a strict honesty discipline. Use it each strategic cycle, right before the
Open skill - /human-summary
Write a human-readable mathematical progress report (compiled PDF) on a project for the operator / the mathematician who posed the problem. This is NOT `elaboration` (dense input for the strategy consult) and NOT the dashboard. Render from the fact graph's verified
Open skill - /initialize
First-run setup interview for a Danus deployment. Run it on the FIRST session, whenever runtime/.danus-initialized is absent or OPERATOR.md is still the blank template, or when the operator asks to set up / initialize / onboard / re-configure. It greets the operator, explains
Open skill - /write-paper
Turn a project's verified fact graph into a publishable LaTeX paper in a configurable house style — a standalone amsart .tex with a real bibliography, compiled to PDF. Use when a project's target theorem is established and the operator wants the paper, or asks to
Open skill - /check-referenced-statements
Validate externally referenced theorems by querying arXiv theorem search first and Codex's built-in web search second. Use when a markdown proof cites statements from external papers.
Open skill

