Claude Skills by panjose

Run the full Co-Scientist pipeline for one research run.

Evaluate whether a hypothesis newly entered the current top-k set and update the convergence counter deterministically.

Assemble a dashboard-ready snapshot from the canonical state.

Choose exactly one concrete evolution strategy for the active evolution round.

Run the assumption decomposition and deep verification review for a hypothesis.

Run the iterative evolution loop from the current persisted run state until convergence.

Generate exactly one child hypothesis that improves logical coherence, causal consistency, and assumption hygiene.

Generate exactly one child hypothesis by combining complementary strengths from multiple parent hypotheses.

Generate exactly one child hypothesis that is more experimentally and operationally feasible than its parent.

Generate exactly one grounded child hypothesis by strengthening evidence, specificity, and literature support.

Generate exactly one cross-parent child hypothesis by transferring a useful principle from one parent context into another.

Generate exactly one divergent but still testable child hypothesis that challenges the shared assumptions of the parent set.

Generate exactly one child hypothesis that preserves the core idea while reducing unnecessary complexity.

Run the full literature-grounded review for a hypothesis.

Generate exactly one hypothesis candidate by enumerating and combining testable assumptions.

Generate exactly one hypothesis candidate through a structured scientific debate.

Generate exactly one literature-grounded hypothesis candidate for the active round.

Dispatch one enabled generation strategy for a given `ResearchPlan`.

Run the initial review gate for a hypothesis.

Evaluate a hypothesis against prior observations.

Judge one placement-tournament matchup between a candidate hypothesis and one opponent.

Update hypothesis proximity state by invoking the canonical embedding bridge for one hypothesis.

Judge one ranked-tournament matchup between two top frontier hypotheses.

Update ranking artifacts for one reviewed hypothesis using canonical placement-opponent selection, ranked-frontier selection, tournament judgments, and Elo updates.

Run the decomposed review pipeline for a single hypothesis and persist each review stage as a structured artifact.

Summarize the completed reviews for a hypothesis.

Simulate the hypothesis mechanism and identify failure scenarios.

Extract recurring critique patterns from the completed review of a hypothesis.

Select the next island strategy and parent hypothesis set for one evolution round.

Run the canonical literature search bridge for one evidence query and persist traceable evidence artifacts.

Apply deterministic Elo updates for one completed tournament batch, persist touched hypotheses, and write the ranking update receipt.

Generate a `ResearchPlan` from `input.md` or equivalent raw research input.

Generate the final research overview from the top-ranked hypotheses.

Interpret the user's high-level research intent into `RUN_POLICY.yaml` and `state/POLICY_DECISION.json`.

Goal: - Resolve the ready dashboard runtime for one Co-Scientist run and return the current run links to the user. Expected input: - one run directory such as `runs/test1` Execution steps: 1. Resolve the run directory and confirm that it contains the run-local dashboard/state artifacts. 2. Run: ```bash python -m tools.host.claude_project_cli dashboard <run-dir> ``` 3. Read: - `runs/<run_id>/dashboard/LINKS.json` - `runs/<run_id>/dashboard/LINKS.md` 4. Read the CLI JSON result and use it as th...

Goal: - Run the Co-Scientist environment diagnostics and explain any failing or warning checks to the user. Expected input: - no arguments for the default project-local doctor Execution steps: 1. Run: ```bash python -m tools.host.claude_project_cli doctor ``` or, when machine-readable output is needed: ```bash python -m tools.host.claude_project_cli doctor --format json ``` 2. Read the emitted diagnostic output and summarize: - Python version health - Python dependency health - active environ...

Goal: - Explain or execute the project-local Co-Scientist skill installation flow for Claude Code users. Expected input: - no arguments for the standard install path - or a user request to reconcile or uninstall the project-local install Execution steps: 1. If the user wants a fresh install, tell them to run one of: `uv` quick install: ```powershell uv sync --extra dev --extra mcp pnpm --dir apps/dashboard install pnpm --dir apps/dashboard build powershell -File tools/install/install_co_scien...

Goal: - Show the user the available high-level start parameters for Co-Scientist. When to use: - The user asks what knobs or parameters are available. - The user asks how `/co-scientist-start` can be configured. - The user wants the expert-facing parameter table instead of the short guided intake. Execution steps: 1. Run: ```bash python -m tools.host.claude_project_cli params ``` 2. If the user explicitly wants machine-readable output, run: ```bash python -m tools.host.claude_project_cli para...

Goal: - Resume one interrupted Co-Scientist run from Claude Code. Expected input: - one run directory such as `runs/test1` Execution steps: 1. Resolve the run directory and confirm that `input.md` and `state/` artifacts exist. 2. Run: ```bash python -m tools.host.claude_project_cli resume <run-dir> --skill co-scientist-pipeline ``` 3. Read the persisted state artifacts: - `runs/<run_id>/state/PIPELINE_STATE.json` - `runs/<run_id>/state/CURRENT_STAGE.json` - `runs/<run_id>/state/HOST_AGENT_HAN...

Goal: - Start one Co-Scientist run from Claude Code and then continue with the canonical pipeline. Expected input: - one run directory such as `runs/test1` - or one compatibility config path such as `runs/test1/config.yaml` Execution steps: 1. Resolve the run directory or compatibility config path relative to the repository root when the user gives a relative path. 2. Run: ```bash python -m tools.host.claude_project_cli run <run-target> --skill co-scientist-pipeline ``` 3. Read the emitted ha...

Goal: - Start one Co-Scientist run from Claude Code using a natural-language goal or an imported brief. Expected input: - either a natural-language research goal - or a brief path such as `notes/drug_resistance_brief.md` - or no explicit arguments, which should trigger a short guided intake Guided intake: - If the user only says `/co-scientist-start`, ask at most four short questions: 1. What is the research goal? 2. Should the run favor exploration or grounded progress? 3. Should the run use...

Goal: - Validate one Co-Scientist run directory from Claude Code. Expected input: - one run directory such as `runs/test1` Execution steps: 1. Resolve the run directory. 2. Decide whether the user wants a fresh-run validation or a resume validation. 3. Run one of: ```bash python -m tools.validation.contract_validation <run-dir> --skill co-scientist-pipeline ``` or ```bash python -m tools.validation.contract_validation <run-dir> --resume --skill co-scientist-pipeline ``` 4. Read and summarize ...

Resolve dashboard links for one Co-Scientist run from Codex.

Diagnose the local Co-Scientist host-agent environment from Codex.

Show the supported Co-Scientist start parameters for Codex users.

Resume an interrupted Co-Scientist run from Codex.

Bootstrap an existing Co-Scientist run directory from Codex.

Start one Co-Scientist run from Codex using a research goal or imported brief.

Validate one Co-Scientist run directory from Codex.

Produce the current `state/STRATEGY_PLAN.json` bundle for configuration, generation, review, evolution, or overview routing.