Humanive / cogfree

# CogFree
Lightweight skills for freeing cognition from working-memory limits.
This collection packages 28 lightweight cognitive skills. The canonical skill instructions are kept in `skills/`; the self-contained Codex and Claude plugin packages are generated under `packages/`.
## Skills
### Understand
- `restate` — restate your intent before continuing.
- `clarify` — ask the single most important clarification question.
- `grill` — probe thinking iteratively through small sets of useful questions.
- `reconstruct` — derive a concept by rebuilding why it had to be invented.
### Inspect
- `state` — recover the current state of our thinking.
- `unknowns` — map what is known, unknown, implicit, or potentially missing.
- `assumptions` — identify what our current thinking depends on.
- `open` — identify important unresolved parts of our thinking.
- `timeline` — reconstruct how our thinking evolved through meaningful phases.
- `diff` — identify how our thinking changed over the conversation.
### Transform
- `abstract` — find the most general useful pattern or principle.
- `ground` — return to concrete mechanisms, examples, or situations.
- `compress` — find the smallest useful representation of our thinking.
- `reframe` — explore a different representation or perspective.
- `compare` — compare competing ideas or interpretations.
- `synthesize` — combine important ideas into one coherent model.
- `transfer` — apply an insight to other problems, domains, or situations.
### Evaluate
- `challenge` — find the strongest counterarguments, counterexamples, or failure cases.
- `prediction` — derive observable predictions from our current understanding.
- `test` — test whether our current understanding holds.
- `verify` — check important claims against available evidence.
### Continue
- `branch` — identify meaningful directions to explore.
- `anchor` — reconnect the discussion to your original intent.
- `checkpoint` — create a resumable checkpoint of our current thinking.
### Execute
- `next` — identify useful candidate next steps.
- `prioritize` — rank questions or branches by expected value for our goal.
- `act` — turn the response into one clear, executable next action.
- `delegate` — turn a request into a clear, verifiable agent brief.
## Install
### Claude Code
Run in Claude Code:
 [code] 
### Codex
Run in your terminal:
 [code] 
### Pi
Run in your terminal:
 [code] 
### Any supported agent
Install the shared skills directly with `npx skills`:
 [code] 
Install only one skill:
 [code] 
The root `skills/` directory is the canonical source. The Claude and Codex plugins use the host-specific layouts under `packages/`. Pi and `npx skills` install directly from the root `skills/` directory. The Pi package does not add executable tools or extensions.
## Marketplace installation
The repository includes marketplace descriptors at `.codex-plugin/marketplace.json` and `.claude-plugin/marketplace.json` for the commands above.