fix(deep-research): return results in ResearchNotes, not write_file

[zhongxuanwang-nv]

Overview

Fixes a contradicted instruction in the deep researcher's research/ sandbox skills.

The researcher worker is told via research.py (format_research_request):

"Do not call write_file or edit_file; run_research_batch will persist the returned
ResearchNotes under /shared/."

…and _persist_research_notes does exactly that automatically. But three skills
instructed the agent to persist artifacts to /shared via write_file —
data-table-analysis, lightweight-calculation, and forecast-analysis — i.e. an
instruction the researcher is explicitly told not to follow.

This reword changes each skill's persistence step to "include the result in your
returned ResearchNotes (e.g. a ResearchFinding's evidence / narrative_notes); do
not call write_file/edit_file." The separate, still-correct rule that sandbox
code uses /workspace and cannot touch /shared is unchanged.

The same conflict exists in two in-flight skills, fixed/flagged separately:
image-processing carries this fix in #285; chart-generation in #280 still needs it
(flagged to its author).

Validation

$ uv run pytest tests/aiq_agent/agents/deep_researcher/test_deepagents_runtime.py -q
18 passed

$ grep -rn "write_file" .../skills/research/   # now only "do not call write_file"

• I ran the relevant local checks or explained why they are not applicable.
• I added or updated tests for behavior changes. (N/A — wording-only in skill docs; discovery/content tests unaffected and pass.)
• I updated documentation for user-facing or contributor-facing changes.
• I confirmed this PR does not include secrets, credentials, or internal-only data.
• I certify this contribution under the DCO and signed my commits with git commit -s.

Where should reviewers start?

research/data-table-analysis/SKILL.md (steps 5) vs tools/research.py:46
(format_research_request) — the reworded persistence step now matches what the
researcher is actually told to do.

Related Issues

• Relates to: AIQ-3362 (deep researcher skills). Companion to feat(deep-research): add image-processing reference skill #285 (image-processing) and feat(deep-research): provider-neutral sandbox + durable artifact runtime #280 (chart-generation).

Summary by CodeRabbit

• Documentation
  • Updated research skill guidance so calculated outputs are returned directly in research notes for later synthesis.
  • Clarified that compact forecast and calculation results should be captured only after successful processing.
  • Removed instructions that required saving generated artifacts to shared file locations.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Caution

Review failed

An error occurred during the review process. Please try again later.

Walkthrough

The research skill docs now direct computed outputs and compact inputs to ResearchNotes instead of /shared files. The updates cover data-table analysis, forecast analysis, and lightweight calculation instructions.

Changes

Research output contract updates

Layer / File(s)	Summary
Data-table analysis output contract src/aiq_agent/agents/deep_researcher/skills/research/data-table-analysis/SKILL.md	The required execution standard, final output flow, and output formats now place markdown/CSV/JSON results in ResearchNotes and remove /shared file-writing guidance.
Forecast and calculation output contract src/aiq_agent/agents/deep_researcher/skills/research/forecast-analysis/SKILL.md, src/aiq_agent/agents/deep_researcher/skills/research/lightweight-calculation/SKILL.md	Forecast inputs and calculation results are now returned in ResearchNotes, with execution steps updated to do so after successful calculations and without /shared artifact writes.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~4 minutes

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

Warning

Tools execution failed with the following error:

Failed to run tools: Ping-pong health check failed

---

_{Comment @coderabbitai help to get the list of available commands.}

[cdgamarose-nv]

Good catch, thank you!

[cdgamarose-nv]

/ok to test 289319e