Determine sub-agent count in pre-script based on PR size

[Copilot]

This change makes PR review fan-out deterministic by deciding sub-agent usage in the pre-script from PR size, then writing explicit runtime instructions for the reviewer to follow.

What changed

1) Pre-script now computes review strategy from file count

In .github/workflows/gh-aw-fragments/pr-context.md, the pre-script now:

• Computes FILE_COUNT from /tmp/pr-context/files.json
• Computes AGENT_COUNT using fixed thresholds
• Writes /tmp/pr-context/agent_count.txt
• Writes /tmp/pr-context/review-strategy.md with exact instructions for one of three modes:
  • Direct review (no sub-agents)
  • 2 sub-agents (A→Z and Z→A)
  • 3 sub-agents (A→Z, Z→A, largest-first)

It also adds agent_count.txt and review-strategy.md to the generated /tmp/pr-context/README.md manifest.

2) Review prompt now consumes precomputed strategy

In .github/workflows/gh-aw-pr-review.md:

• Step 2 is simplified to read and follow /tmp/pr-context/review-strategy.md (instead of hard-coding 3 sub-agents)
• Step 3 now states dedup/merge is conditional on sub-agents being used

Thresholds

Files changed	Sub-agents	Strategy
<= 10	0	Parent agent reviews directly
11-20	2	A→Z + Z→A
> 20	3	A→Z + Z→A + largest-first

Generated workflow updates

• .github/workflows/gh-aw-pr-review.lock.yml includes the compiled logic/prompt updates above
• Other lock files that import the updated fragment were regenerated and only had metadata hash refreshes

Fixes elastic/ai-github-actions#508.

Generated by Update PR Body for issue #510

[coderabbitai]

📝 Walkthrough

Walkthrough

This change introduces dynamic sub-agent sizing for PR reviews based on the number of changed files. The PR context generation now calculates a FILE_COUNT and assigns AGENT_COUNT values (0 for ≤10 files, 2 for 11–20 files, 3 for >20 files), writing these values to agent_count.txt. A new review-strategy.md artifact is generated with conditional strategy sections for Direct Review (no sub-agents), 2-agent review, or 3-agent review depending on PR size. Workflow files are updated to consume this strategy document instead of hard-coded instructions, and the PR context manifest is expanded to include these new precomputed artifacts.

Possibly related PRs

• Updates to Workflows for subagents #445: Modifies PR review workflows and sub-agent orchestration logic; this PR extends the same pr-context/sub-agent flow by adding size-driven agent count and strategy artifacts.
• Add pre-run file creation for unresolved, resolved, and outdated threads #503: Adds new precomputed PR-context artifacts and updates workflow prompts to consume them, following a similar pattern of artifact generation and manifest updates.
• Instruct agents not to mention on-disk data sources in responses #499: Modifies PR context fragments and on-disk artifacts; this PR introduces new /tmp/pr-context files (agent_count.txt, review-strategy.md) that align with artifact management practices discussed there.

🚥 Pre-merge checks | ✅ 2

✅ Passed checks (2 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	PR implements all core requirements from issue #508: computes PR size in prescript, applies three-tier thresholds (≤10, 11-20, >20 files), generates deterministic strategy instructions, and integrates with review workflow.
Out of Scope Changes check	✅ Passed	All changes align with issue #508 objectives: pre-script sizing logic, strategy document generation, manifest updates, and review workflow alignment. No unrelated modifications detected.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch copilot/determine-agent-count

---

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

[github-actions]

[HIGH] Conflicting instructions still force three sub-agents

This new strategy step says to follow /tmp/pr-context/review-strategy.md (which can require 0 or 2 agents), but the workflow still imports gh-aw-fragments/pick-three-keep-many.md, whose text explicitly says “Spawn 3 sub-agents” and “Wait for all 3 sub-agents”.

Because both instructions are present in the same prompt, the model can still follow the hardcoded “3 sub-agents” rule, which defeats the deterministic sizing behavior introduced here.

Please make the prompt sources consistent (for example, gate/remove the pick-three-keep-many fragment when strategy says direct review, or rewrite that fragment to be conditional on review-strategy.md).

[strawgate]

/ai resolve pr issues

[coderabbitai]

♻️ Duplicate comments (1)

.github/workflows/gh-aw-pr-review.md (1)

125-125: ⚠️ Potential issue | 🟠 Major

Resolve conflicting sub-agent instructions in the prompt.

Line 125 makes orchestration dynamic, but this prompt still imports gh-aw-fragments/pick-three-keep-many.md (Line 16). If that fragment still contains fixed 3-agent orchestration text, the agent receives contradictory directives and deterministic PR-size fan-out can fail.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In @.github/workflows/gh-aw-pr-review.md at line 125, The prompt imports a fixed
3-agent fragment (gh-aw-fragments/pick-three-keep-many.md) while orchestration
is supposed to be dynamic via the pre-computed review strategy (read
`/tmp/pr-context/review-strategy.md`), causing conflicting directives; update
the prompt assembly in gh-aw-pr-review.md so the pick-three-keep-many fragment
is only included when the strategy explicitly requires three sub-agents, or
replace the fixed fragment import with logic that reads
`/tmp/pr-context/review-strategy.md` and conditionally includes the appropriate
fragment (0, 2, or 3-agent fragment names) to ensure the chosen orchestration is
deterministic and consistent with the review strategy.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In @.github/workflows/gh-aw-pr-review.md:
- Line 125: The prompt imports a fixed 3-agent fragment
(gh-aw-fragments/pick-three-keep-many.md) while orchestration is supposed to be
dynamic via the pre-computed review strategy (read
`/tmp/pr-context/review-strategy.md`), causing conflicting directives; update
the prompt assembly in gh-aw-pr-review.md so the pick-three-keep-many fragment
is only included when the strategy explicitly requires three sub-agents, or
replace the fixed fragment import with logic that reads
`/tmp/pr-context/review-strategy.md` and conditionally includes the appropriate
fragment (0, 2, or 3-agent fragment names) to ensure the chosen orchestration is
deterministic and consistent with the review strategy.

---

ℹ️ Review info

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 4e93b38 and 7b2304a.

📒 Files selected for processing (9)

• .github/workflows/gh-aw-fragments/pr-context.md
• .github/workflows/gh-aw-mention-in-pr-by-id.lock.yml
• .github/workflows/gh-aw-mention-in-pr-no-sandbox.lock.yml
• .github/workflows/gh-aw-mention-in-pr.lock.yml
• .github/workflows/gh-aw-pr-review-addresser.lock.yml
• .github/workflows/gh-aw-pr-review.lock.yml
• .github/workflows/gh-aw-pr-review.md
• github/workflows/gh-aw-fragments/pick-three-keep-many.md
• github/workflows/gh-aw-pr-review.md