Edison

Edison Research

You are delegating a scientific literature search to the Edison API. Edison is an autonomous research agent that searches hundreds of papers and synthesizes a cited answer. Your job is to submit the query, wait for the result, and report it.

CRITICAL: Do NOT search for the answer yourself

Edison takes 1–5 minutes to respond. This is normal. BE PATIENT.

While waiting for Edison, you MUST NOT:

• Search the web, PubMed, bioRxiv, or any other source
• Attempt to answer the research question from your own knowledge
• Use any other tool to find papers or literature
• Speculate about what the answer might be

Edison's search is comprehensive (hundreds of papers with full-text analysis). Anything you find on your own would be redundant and inferior. Your role is to submit and wait, not to compete with Edison.

The ONLY thing you may do while an Edison query is running is launch additional Edison queries if the user asked for multiple.

Input

If no query is provided, ask the user what they want to research.

Step 1: Run the query

Resolve the Edison script path first:

• Claude Code exposes the skill directory as CLAUDE_SKILL_DIR.
• In Codex, find the installed SoftNano Edison skill under CODEX_HOME or ~/.codex.

Run the script with run_in_background: true and timeout: 600000 (10 minutes):

if [ -n "${CLAUDE_SKILL_DIR:-}" ]; then
    EDISON_SCRIPT="$CLAUDE_SKILL_DIR/scripts/edison_query.sh"
else
    EDISON_SCRIPT="$(
        find "${CODEX_HOME:-$HOME/.codex}" \
            -path "*/softnanolab-plugins/softnano/*/skills/edison/scripts/edison_query.sh" \
            -print -quit
    )"
fi

if [ -z "${EDISON_SCRIPT:-}" ] || [ ! -x "$EDISON_SCRIPT" ]; then
    echo '{"error": "Could not locate executable edison_query.sh in the installed SoftNano skill."}'
else
    "$EDISON_SCRIPT" --query "<the research question>" --max-wait 540
fi

The script authenticates, submits the query, and polls every 15s until Edison returns an answer. Progress messages appear on stderr. The final result is a JSON object on stdout:

{
  "task_id": "uuid",
  "status": "success",
  "formatted_answer": "The full cited answer in markdown",
  "answer": "Plain text answer",
  "has_successful_answer": true
}

If the local wait limit is reached before Edison finishes, the task is still running on Edison's servers. The script returns a resumable, non-fatal JSON object:

{
  "task_id": "uuid",
  "status": "running",
  "recoverable": true,
  "has_successful_answer": false,
  "resume_command": "/path/to/edison_query.sh --task-id uuid"
}

In this case, do not answer from memory or search other sources. Poll the same task id again:

"$EDISON_SCRIPT" --task-id <task_id> --max-wait 540

Repeat polling until status is success or Edison returns a terminal failure. If the shell tool itself times out but the last visible output contains a task_id, recover using --task-id <task_id>.

If the script outputs an error field, report the error to the user and stop.

API key not set: If the error indicates EDISON_PLATFORM_API_KEY not set, immediately ask the user to provide their Edison API key. Tell them they can get one from platform.edisonscientific.com/profile and set it in their agent environment. Claude Code users can put it in ~/.claude/settings.json under "env"; Codex users can export it in their shell or project environment. Do NOT proceed with any other action — wait for the user to provide the key.

Running multiple queries

If the user asks multiple research questions, launch each as a separate background Bash call. Wait for all to complete before reporting.

Step 2: Report the answer

Once the script returns:

1. Report formatted_answer to the user verbatim — it contains inline citations and is the primary output.
2. Note the task_id — the user may want it for follow-up queries.
3. If has_successful_answer is false, tell the user Edison could not confidently answer and suggest refining the question.

Follow-up queries

To continue a previous research thread (e.g., ask a follow-up question that builds on prior context), use --continue-from:

if [ -n "${CLAUDE_SKILL_DIR:-}" ]; then
    EDISON_SCRIPT="$CLAUDE_SKILL_DIR/scripts/edison_query.sh"
else
    EDISON_SCRIPT="$(
        find "${CODEX_HOME:-$HOME/.codex}" \
            -path "*/softnanolab-plugins/softnano/*/skills/edison/scripts/edison_query.sh" \
            -print -quit
    )"
fi

if [ -z "${EDISON_SCRIPT:-}" ] || [ ! -x "$EDISON_SCRIPT" ]; then
    echo '{"error": "Could not locate executable edison_query.sh in the installed SoftNano skill."}'
else
    "$EDISON_SCRIPT" --query "<follow-up question>" --continue-from <task_id> --max-wait 540
fi

This reuses the papers and context from the previous task, giving Edison a head start. Use this when the user wants to dig deeper into a topic they already queried.

Prerequisites

The user must set EDISON_PLATFORM_API_KEY in the agent environment. Claude Code users can set it in ~/.claude/settings.json:

{
  "env": {
    "EDISON_PLATFORM_API_KEY": "your-key-here"
  }
}

Keys are available from platform.edisonscientific.com/profile.

jq must be installed (brew install jq / apt install jq).