panjose/Co-Scientist / skills/skills-codex-entry/co-scientist-start
co-scientist-start
VerifiedStart one Co-Scientist run from Codex using a research goal or imported brief.
co-scientist-start
Goal:
- Create a Co-Scientist run, bootstrap the canonical host-agent pipeline, and continue from the repository workflow skills.
Expected input:
- a natural-language research goal
- or a brief path
- or no explicit arguments, which should trigger a short guided intake
Execution steps:
-
If the user only invokes
$co-scientist-start, ask only for the missing high-level run inputs: goal, exploration preference, iteration policy, and optional brief path. -
Convert high-level controls into explicit CLI flags. Do not pass free-form
key: valuetext when a real flag exists. -
Before creating files, run:
python -m tools.host.project_cli start --goal "<goal>" --skill co-scientist-pipeline --summary-onlyUse the equivalent
--brief <path>command for imported briefs. -
Show the returned summary and wait for user confirmation.
- Tell the user that the run-local dashboard receipt will be written to:
runs/<run_id>/dashboard/LINKS.mdruns/<run_id>/dashboard/LINKS.json
- Tell the user that
$co-scientist-dashboard <run-dir>is the ready-link follow-up when the background bootstrap has not finished yet.
- Tell the user that the run-local dashboard receipt will be written to:
-
After confirmation, run:
python -m tools.host.project_cli start --goal "<goal>" --skill co-scientist-pipeline -
Read the emitted
runs/<run_id>/state/HOST_AGENT_HANDOFF.json. -
Read the CLI JSON result and the run-local dashboard receipt artifacts:
runs/<run_id>/dashboard/LINKS.mdruns/<run_id>/dashboard/LINKS.json
-
If the CLI JSON contains
dashboardLinks:- If
dashboard.statusisrunning, returndashboardLinks.dashboardas the primary dashboard URL and include the deep links. - If
dashboard.statusisstarting, immediately run:
python -m tools.host.project_cli dashboard <run-dir>- Read the refreshed CLI JSON result plus
runs/<run_id>/dashboard/LINKS.md. - If
runtime.statusis nowrunning, return the refreshedlinks.dashboardURL as the primary dashboard URL and include the deep links. - If
runtime.statusis stillstarting, tell the user that the dashboard is still booting, point them toruns/<run_id>/dashboard/LINKS.md, and include the retry command:
$co-scientist-dashboard <run-dir> - If
-
Open
skills/co-scientist-pipeline/SKILL.mdand the listed shared references from the handoff. -
Continue execution from the canonical repository-local workflow. If routing returns
run_configuration, runresearch-configand validateresearch_plan/RESEARCH_PLAN.jsonbefore generation. -
After major phase writes, run:
python -m tools.validation.contract_validation runs/<run_id> --skill co-scientist-pipeline
Rules:
- Use
tools.host.project_cli; do not call the Claude-specific CLI from Codex entry skills. - Treat
skills/as canonical and.agents/skills/as the installed Codex discovery surface. - Do not invent run artifacts. Read the CLI JSON and run-local receipts.
- Treat
runs/<run_id>/dashboard/LINKS.mdas the human-readable dashboard receipt andruns/<run_id>/dashboard/LINKS.jsonas the machine-readable receipt. - Fresh runs normally pass through the explicit
Configurationstage first soresearch-configcan materializeresearch_plan/RESEARCH_PLAN.json. - Under
completion_driven + auto, continue until the routing plan reaches overview, a configured checkpoint, or a blocking validator/safety state. - If you must stop before convergence or a terminal route, tell the user the run is paused, current convergence has not been reached, persisted state is resumable, and the next recommended action is continue evolution through
$co-scientist-resume <run-dir>or an explicit continue request.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |