allenai Research Challenge
Capture what a research project in the current working directory accomplished, how the Asta skills contributed, and (optionally) submit the project as a contribution to https://github.com/allenai/asta-research-challenge.
This skill is a router. Pick the workflow that matches the user's intent, open its .md file in workflows/, and follow it. Do not execute a workflow from memory.
Files
| Path | Role |
|---|---|
<cwd>/RESEARCH_CHALLENGE.md |
Output of reflect. Project summary, self-critique, and skill-improvement suggestions. Input to submit. |
~/.claude/projects/<encoded-cwd>/ |
Source of conversation history transcripts (JSONL) — copied by submit. |
The <encoded-cwd> is the current working directory with / replaced by - (e.g. /Users/x/work/proj → -Users-x-work-proj).
Workflows
| Name | Purpose | Detailed instructions |
|---|---|---|
| reflect | Interview the user about the project, self-critique the agent's contribution, suggest skill improvements, and write RESEARCH_CHALLENGE.md. |
workflows/reflect.md |
| submit | Copy the report, conversation transcripts, and project artifacts into a new project-specific directory in a clone of allenai/asta-research-challenge, then open a PR. |
workflows/submit.md |
Routing
- Explicit request. If the user names a workflow ("reflect on this", "write the report", "submit the challenge", "publish it"), dispatch directly.
- Implicit "wrap up" intent. If the user is winding down a project but hasn't named a workflow, default to reflect, then offer submit at the end.
- Submit without a report. If the user asks to submit but no
RESEARCH_CHALLENGE.mdexists, run reflect first and chain to submit on confirmation.
Out of scope
- Grading research quality. The reflect workflow produces an honest self-critique, not a peer review.
- Pushing to
mainor merging the PR. submit opens a PR and stops.