exploreAIExplore
How to Use AdaL for Deep research subagents
Learn AdaL deep research subagents with step by step workflows, realistic examples, and verified plan notes.
This deep research subagents guide shows a practical AdaL path from brief to reviewable output. Lead with CI test pattern, use commit message, and keep test driven secondary until the core result is right. Plans: adalagent.ai. Explore: /explore/adal.
Below is a full deep research subagents walkthrough. See also /blog/how-to-use-adal-for-code-review-workflows, /blog/how-to-use-adal-for-refactoring-with-constraints, /blog/how-to-use-adal-for-documentation-updates.
When this workflow is the right job
Pick deep research subagents for a focused AdaL pass. Skip it when cli coding tasks with tests or code review workflows covers the requirement more directly.
Step by step workflow
1. Brief Deep research subagents
Write what must stay true for deep research subagents in AdaL before settings or spend.
Brief: Deep research subagents Keep: refactor validateEmail from SOURCE Avoid: invented pricing or features Success: one reviewable output
2. Open AdaL for Deep research subagents
Use the AdaL surface that owns deep research subagents. Do not mix a neighboring workflow in the same pass.
Surface: Deep research subagents Start: worktree isolate Plans: adalagent.ai
3. Pilot Deep research subagents
Run a single deep research subagents pilot. Score clarity, grounding, and whether CI strict still matches.
Pilot: Deep research subagents [ ] SOURCE facts match [ ] symbol scoped clear [ ] Settings logged
4. Refine Deep research subagents
Change one deep research subagents dimension only. Save a template with variables for lint cleanup.
Refine: Deep research subagents Change: subagent route Keep: SOURCE and safety constrained
Practical deep research subagents examples
refactor validateEmail
Scenario: A developer uses AdaL for deep research subagents scoped to "refactor validateEmail". Objective: Make a minimal, test-backed change that satisfies refactor validateEmail without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for refactor validateEmail - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for refactor validateEmail. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing refactor validateEmail, with test results and a short file list.
README install
Scenario: A developer uses AdaL for deep research subagents scoped to "README install". Objective: Make a minimal, test-backed change that satisfies README install without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for README install - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for README install. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing README install, with test results and a short file list.
bug repro
Scenario: A developer uses AdaL for deep research subagents scoped to "bug repro". Objective: Make a minimal, test-backed change that satisfies bug repro without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for bug repro - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for bug repro. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing bug repro, with test results and a short file list.
feature branch
Scenario: A developer uses AdaL for deep research subagents scoped to "feature branch". Objective: Make a minimal, test-backed change that satisfies feature branch without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for feature branch - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for feature branch. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing feature branch, with test results and a short file list.
CI test pattern
Scenario: A developer uses AdaL for deep research subagents scoped to "CI test pattern". Objective: Make a minimal, test-backed change that satisfies CI test pattern without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for CI test pattern - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for CI test pattern. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing CI test pattern, with test results and a short file list.
API freeze
Scenario: A developer uses AdaL for deep research subagents scoped to "API freeze". Objective: Make a minimal, test-backed change that satisfies API freeze without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for API freeze - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for API freeze. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing API freeze, with test results and a short file list.
RFC compare
Scenario: A developer uses AdaL for deep research subagents scoped to "RFC compare". Objective: Make a minimal, test-backed change that satisfies RFC compare without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for RFC compare - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for RFC compare. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing RFC compare, with test results and a short file list.
symbol scoped
Scenario: A developer uses AdaL for deep research subagents scoped to "symbol scoped". Objective: Make a minimal, test-backed change that satisfies symbol scoped without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for symbol scoped - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for symbol scoped. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing symbol scoped, with test results and a short file list.
type error
Scenario: A developer uses AdaL for deep research subagents scoped to "type error". Objective: Make a minimal, test-backed change that satisfies type error without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for type error - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for type error. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing type error, with test results and a short file list.
flaky test
Scenario: A developer uses AdaL for deep research subagents scoped to "flaky test". Objective: Make a minimal, test-backed change that satisfies flaky test without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for flaky test - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for flaky test. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing flaky test, with test results and a short file list.
migration script
Scenario: A developer uses AdaL for deep research subagents scoped to "migration script". Objective: Make a minimal, test-backed change that satisfies migration script without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for migration script - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for migration script. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing migration script, with test results and a short file list.
lint cleanup
Scenario: A developer uses AdaL for deep research subagents scoped to "lint cleanup". Objective: Make a minimal, test-backed change that satisfies lint cleanup without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for lint cleanup - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for lint cleanup. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing lint cleanup, with test results and a short file list.
dependency bump
Scenario: A developer uses AdaL for deep research subagents scoped to "dependency bump". Objective: Make a minimal, test-backed change that satisfies dependency bump without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for dependency bump - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for dependency bump. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing dependency bump, with test results and a short file list.
perf hotspot
Scenario: A developer uses AdaL for deep research subagents scoped to "perf hotspot". Objective: Make a minimal, test-backed change that satisfies perf hotspot without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for perf hotspot - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for perf hotspot. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing perf hotspot, with test results and a short file list.
security review
Scenario: A developer uses AdaL for deep research subagents scoped to "security review". Objective: Make a minimal, test-backed change that satisfies security review without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for security review - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for security review. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing security review, with test results and a short file list.
docs sync
Scenario: A developer uses AdaL for deep research subagents scoped to "docs sync". Objective: Make a minimal, test-backed change that satisfies docs sync without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for docs sync - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for docs sync. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing docs sync, with test results and a short file list.
worker agent
Scenario: A developer uses AdaL for deep research subagents scoped to "worker agent". Objective: Make a minimal, test-backed change that satisfies worker agent without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for worker agent - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for worker agent. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing worker agent, with test results and a short file list.
patch release
Scenario: A developer uses AdaL for deep research subagents scoped to "patch release". Objective: Make a minimal, test-backed change that satisfies patch release without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for patch release - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for patch release. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing patch release, with test results and a short file list.
smoke suite
Scenario: A developer uses AdaL for deep research subagents scoped to "smoke suite". Objective: Make a minimal, test-backed change that satisfies smoke suite without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for smoke suite - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for smoke suite. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing smoke suite, with test results and a short file list.
auth test fix
Scenario: A developer uses AdaL for deep research subagents scoped to "auth test fix". Objective: Make a minimal, test-backed change that satisfies auth test fix without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for auth test fix - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for auth test fix. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing auth test fix, with test results and a short file list.
Deep Research memo
Scenario: A developer uses AdaL for deep research subagents scoped to "Deep Research memo". Objective: Make a minimal, test-backed change that satisfies Deep Research memo without drive-by refactors or public API breaks. Inputs: - Failing test or reproduction for Deep Research memo - File/path scope - Repo CI command - API freeze constraints Workflow: Read failing test → Plan minimal diff → Edit → Run CI command → Summarize files touched Requirements: - Minimal diff; no unrelated refactors. - Do not change public API signatures unless required for Deep Research memo. - Run the documented test/CI command. - Do not invent secrets or environment values. Expected output: A reviewable patch for deep research subagents addressing Deep Research memo, with test results and a short file list.
How to improve deep research subagents
Make deep research subagents easier to review by labeling README install fields that must never change in AdaL.
Speed deep research subagents iteration by cloning the last good AdaL run and altering only worktree isolate.
Stabilize deep research subagents by pinning minimal diff after feature branch is approved in AdaL.
Reduce deep research subagents rework by rejecting drafts that invent claims about CI test pattern in AdaL.
Improve deep research subagents handoffs by recording which AdaL control produced the API freeze result.
Strengthen deep research subagents by adding a second reader who only checks RFC compare spelling and facts in AdaL.
Lift deep research subagents consistency by reusing the same run tests vocabulary across related AdaL jobs.
Harden deep research subagents by testing an empty or incomplete type error input before trusting AdaL defaults.
Prompting and usage guidance
Lead deep research subagents with constraints: channel, length, and forbidden claims inside AdaL.
Separate creative instructions from SOURCE so deep research subagents stays grounded in AdaL.
Request deep research subagents output as a checklist first when stakeholders need approval gates.
For deep research subagents, describe feature branch with concrete nouns, then add research cited only if the draft already works.
Ask AdaL to list assumptions made during deep research subagents before you accept the draft.
Limitations to respect
Do not invent credit costs for deep research subagents; read live numbers on adalagent.ai.
AdaL can be wrong. Treat deep research subagents as provisional until review.
Connected apps used in deep research subagents may throttle traffic independently of AdaL.
If documentation is silent on a deep research subagents claim, leave it out rather than guessing.
Practical tips for this workflow
Document AdaL UI labels used for deep research subagents so handoffs about lint cleanup do not rely on memory.
Store winning deep research subagents settings as a template with variables only for smoke suite fields in AdaL.
Approve SOURCE facts before spending budget on deep research subagents variants that mention CI test pattern in AdaL.
Pair customer facing deep research subagents exports with a human read that checks invented claims about lint cleanup.
Log AdaL run identifiers for deep research subagents so ops can replay read first failures without guessing.
Split oversized deep research subagents work into smaller CI green passes rather than one overloaded AdaL request.
Review deep research subagents while context is fresh; delayed checks miss docs clear mismatches on lint cleanup.
If deep research subagents touches compliance language about smoke suite, lock verbatim strings outside AdaL first.
Retire deep research subagents templates when AdaL docs change names or gates for CI test pattern workflows.
For deep research subagents, capture a before and after artifact of lint cleanup every time AdaL settings change.
AdaL deep research subagents note: after CI green, recheck migration script against SOURCE and confirm minimal diff still matches the brief.
AdaL deep research subagents note: after CLI goal, recheck lint cleanup against SOURCE and confirm dev terse still matches the brief.
AdaL deep research subagents note: after run tests, recheck dependency bump against SOURCE and confirm research cited still matches the brief.
AdaL deep research subagents note: after read first, recheck perf hotspot against SOURCE and confirm safety constrained still matches the brief.
AdaL deep research subagents note: after no public API change, recheck security review against SOURCE and confirm test driven still matches the brief.
AdaL deep research subagents note: after subagent route, recheck docs sync against SOURCE and confirm docs clear still matches the brief.
AdaL deep research subagents note: after cite official docs, recheck worker agent against SOURCE and confirm review friendly still matches the brief.
Common mistakes
- Starting deep research subagents without SOURCE facts in AdaL
- Treating marketing blogs as official AdaL limits
- Regenerating everything when one deep research subagents section failed
- Leaving credentials in deep research subagents node fields instead of vaults
- Promising delivery dates before checking AdaL plan access
- Skipping the human read on customer facing deep research subagents drafts
After this deep research subagents guide, continue with /blog/how-to-use-adal-for-code-review-workflows, /blog/how-to-use-adal-for-refactoring-with-constraints, /blog/how-to-use-adal-for-documentation-updates. Start again at /explore/adal if you need the full AdaL map.