biomni_plus

Imported from x1jiang/biomni_plus (AGENTS.md). Install upstream with npx skills add x1jiang/biomni_plus. Copyright stays with the author.

Failure lessons (2026-08-27)

• Treat the first execute failure (exit code 2) as a logic/invocation problem, not a format error; inspect the command, path, and interpreter before retrying.
• After writing the script, execute it with the exact saved path and a known Python interpreter; the corrected run produced the expected AUGGGCAAGUAA and MGK.
• Preserve the efficient write-then-run workflow, but avoid spending a turn on an unverified command; validate the script path and invocation immediately.
• The objective verified score of 0.80 reflects the successful tool output, while the lower self-assessed run score (1/3) appropriately reflects the mid-run failure and unnecessary recovery; correct results do not make the trajectory clean.

Failure lessons (2026-08-27)

• Use the workspace path returned or permitted by the tool (/Users/xiaoqianjiang/...) instead of guessing /Users/xiaoqi; the initial failure was a path/strategy logic error, not a format error.
• After a path rejection, inspect or use the active workspace directory before retrying file creation.
• The eventual script-and-execute approach worked: transcription produced AUGGGCAAGUAA, translation produced MGK*, and removing the stop marker yielded peptide MGK.
• Explicitly verify both the exit code and biological output; the objective score of 0.80 reflects successful execution with medium-confidence tool verification, while the self-assessed 1/3 reflects the avoidable failed first attempt.