Add deep research workflow

[strawgate]

Let's look at what gh-aw uses for deep research but my guess is we need web search which we can't currently use

[github-actions]

TL;DR: I recommend adding a new installable deep-research workflow (issue-comment driven) rather than reworking internals; the repo already has the needed research tools (web_fetch + search_code), and there is already an internal deep-dive workflow you can reuse as a pattern.

Recommendation

Create a publishable workflow pair:

1. .github/workflows/gh-aw-deep-research.md (workflow definition)
2. gh-agent-workflows/deep-research/{example.yml,README.md} (consumer trigger/docs)

Base it on gh-aw-plan.md + gh-aw-mention-in-issue.md, keep outputs to add-comment (and optionally create-issue), and trigger on issue comments like /ai research ....

Findings

• There is already an internal deep-dive workflow:
  • .github/workflows/agent-deep-dive.md:3-4 (name: "Agent Deep Dive", deep behavioral analysis)
  • .github/workflows/agent-deep-dive.md:44 enables web-fetch
• It is not currently installable as a published gh-aw-* workflow:
  • no .github/workflows/gh-aw-agent-deep-dive.md
  • no trigger-agent-deep-dive.yml (glob returned no matches)
  • no gh-agent-workflows/agent-deep-dive/* directory
• Existing issue-facing patterns to reuse:
  • .github/workflows/gh-aw-issue-triage.md:3-4, 60-64
  • .github/workflows/gh-aw-plan.md:3-4, 61-66
  • gh-agent-workflows/mention-in-issue/example.yml:3-4,15-18 (issue_comment + /ai activation)
• Tooling for research is already available in current workflow architecture:
  • .github/workflows/gh-aw-fragments/elastic-tools.md:5-11 (generate_agents_md, search_code)
  • widespread web-fetch in workflow shims (e.g. .github/workflows/gh-aw-issue-triage.md:64)
• Related prior work:
  • [agent-efficiency] Recurring PR checkout failures in agent workflows (missing remote refs) #307 (requested behavior-focused analysis)
  • PR Overhaul agent-efficiency reporting and add per-agent deep-dive workflow #311 / PR Add agent-deep-dive and agent-efficiency as internal workflows with log analysis scripts #313 (introduced and internalized agent-deep-dive + agent-efficiency)

Verification

I validated workflow baseline and current availability:

$ make lint-workflows && ls -1 .github/workflows | grep -E 'deep|triage|research|agent-efficiency'
...
agent-deep-dive.lock.yml
agent-deep-dive.md
agent-efficiency.lock.yml
agent-efficiency.md
gh-aw-issue-triage.lock.yml
gh-aw-issue-triage.md
trigger-issue-triage.yml

This confirms deep-dive exists internally, while issue-triage is the installable/public pattern.

Detailed Action Plan

1. Create workflow shim: add .github/workflows/gh-aw-deep-research.md by adapting:
   • prompt/process from .github/workflows/gh-aw-plan.md:84-129
   • issue-comment request handling style from .github/workflows/gh-aw-mention-in-issue.md:91-139
2. Define trigger example: add gh-agent-workflows/deep-research/example.yml modeled on gh-agent-workflows/mention-in-issue/example.yml:1-20, with a stricter condition (e.g., startsWith /ai research).
3. Add docs: add gh-agent-workflows/deep-research/README.md following gh-agent-workflows/issue-triage/README.md:1-36 structure.
4. Register docs/listings: update gh-agent-workflows/README.md specialized/default sections (:21-73) so users can discover the workflow.
5. Optional CI coverage: add deep-research to .github/workflows/smoke-test-install.yml:44-81 install matrix.
6. Validate: run make compile then make lint-workflows.

Related Items

Type	Link	Relevance
Issue	#336	Current request
Issue	#307	Prior request for deeper workflow behavior analysis
PR	#311	Added deep-dive concept initially
PR	#313	Merged internal agent-deep-dive/agent-efficiency
File	.github/workflows/agent-deep-dive.md	Existing internal deep analysis workflow
File	.github/workflows/gh-aw-plan.md	Best reusable template for research-style responses
File	gh-agent-workflows/mention-in-issue/example.yml	Trigger pattern for issue-comment activation

---

What is this? | From workflow: Issue Triage

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.

[strawgate]

/ai let's make a PR and let's use claude as the engine and add the web_search tool

[github-actions]

Implemented and submitted a PR payload for a new Deep Research workflow using Claude with web search enabled.

What changed

• Added workflow source scaffold at github/workflows/gh-aw-deep-research.md with engine.id: claude and tools web-search + web-fetch (github/workflows/gh-aw-deep-research.md:16-19, :62-67).
• Added trigger example using /ai research and ANTHROPIC_API_KEY (gh-agent-workflows/deep-research/example.yml:15-20).
• Added workflow docs (gh-agent-workflows/deep-research/README.md:1-41).
• Registered it in the workflow catalog (gh-agent-workflows/README.md:38-40).

Validation

• make lint-workflows passed.
• make compile passed.

Important note

• Per workflow-edit guardrails in this run, I did not edit .github/workflows/ directly; the new workflow source is in github/workflows/gh-aw-deep-research.md and should be relocated by a maintainer to .github/workflows/gh-aw-deep-research.md, then compiled to produce gh-aw-deep-research.lock.yml.

---

What is this? | From workflow: Mention in Issue

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.