feat(backlog-manager): maximize throughput by filling all available capacity slots

[aaight]

Summary

Fixes the backlog-manager agent so it fills ALL remaining capacity slots rather than conservatively moving only one item at a time.

Previously, with maxInFlightItems=2 and two independent backlog items, the agent would move only one to TODO. This PR strengthens the prompt language to eliminate the ambiguity.

• Strengthened system prompt (backlog-manager.eta): rewrites the CAPACITY LIMIT section and Step 3/4 to explicitly instruct the agent to move the maximum number of eligible items, adds a dedicated MAXIMIZE THROUGHPUT rule for maxInFlightItems > 1
• Strengthened YAML task prompt (backlog-manager.yaml): task Step 2 now says "select ALL eligible unblocked items to fill remaining capacity completely — always move the maximum number possible"
• availableSlots added to PipelineCapacityResult: computed in all return paths (backlog-empty = limit, at-capacity = 0, below-capacity = limit − inFlightCount), exposed in all log payloads at call sites in pr-merged.ts and agent-execution.ts
• Tests updated: 26 backlog-check tests now assert availableSlots; 3 new prompt tests verify the strengthened language; worker-entry.test.ts gets a beforeEach fix for CASCADE process environment inheritance

Test plan

• tests/unit/triggers/shared/backlog-check.test.ts — 26 tests all pass, including new availableSlots assertions for every reason code
• tests/unit/agents/prompts.test.ts — 109 tests pass; 3 new tests verify the maximize-throughput language
• tests/unit/worker-entry.test.ts — 21 tests pass (fixed pre-existing failure caused by JOB_* env vars inherited from CASCADE worker process)
• Lint: biome check . — no issues
• Type check: tsc --noEmit — zero errors

Trello card

https://trello.com/c/69ea85be987570b9ab179ebc

🤖 Generated with Claude Code

🕵️ claude-code · claude-sonnet-4-6 · run details

[codecov]

Codecov Report

❌ Patch coverage is 94.11765% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/triggers/shared/agent-execution.ts	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[nhopeatall]

Summary

LGTM - The prompt strengthening provides clear, disambiguated instructions to the agent to ensure maximum throughput, and the addition of availableSlots to the capacity check makes the state clearer in logs.

The test coverage is thorough and perfectly matches the new logic. The fix in worker-entry.test.ts to clear JOB_* env vars is a nice addition that will prevent CI/local environment flakiness.

🕵️ opencode · google/gemini-3.1-pro-preview · run details