Reduce ambient context cost in daily workflows (go-logger, sergo, smoke-copilot)

[Copilot]

Daily optimizer findings showed token burn was dominated by turn proliferation (go-logger), structural prompt overhead (sergo), and verbose static requirements (smoke-copilot), not just raw prompt size. This PR reduces recurring context load by front-loading deterministic preprocessing and removing always-on prompt bulk.

• Go Logger: deterministic preflight + manifest-first flow

  • Added a pre-activation manifest step in go-logger.md to compute:
    • preflight.json (should_run, SHAs, paths)
    • manifest.json (files needing logger/imports/candidate call sites)
  • Gated agent execution on preflight output so unchanged runs can short-circuit.
  • Replaced iterative discovery guidance with manifest consumption.
  • Added compact structured patch schema guidance for logger diffs.

• Sergo: flatten prompt structure + externalize examples

  • Refactored sergo.md from heading-heavy/template-heavy form to a compact execution plan.
  • Removed embedded example fences from the base prompt.
  • Added bounded strategy-history loading guidance (recent entries only).
  • Introduced conditional examples skill: .github/skills/sergo-examples/SKILL.md.

• Smoke Copilot: trim largest static section

  • Condensed ## Test Requirements in smoke-copilot.md while preserving required checks and outputs.
  • Reduced repeated phrasing/instruction overhead in the highest-cost section.

• Compiled artifacts updated

  • Regenerated:
    • .github/workflows/go-logger.lock.yml
    • .github/workflows/sergo.lock.yml
    • .github/workflows/smoke-copilot.lock.yml

Example (go-logger preflight gate + manifest output):

steps:
  - name: Build deterministic logger manifest
    id: logger_manifest
    run: |
      # writes /tmp/gh-aw/agent/go-logger/preflight.json + manifest.json
      # emits should_run=true|false
if: needs.pre_activation.outputs.should_run == 'true' || github.event_name == 'workflow_dispatch'
jobs:
  pre-activation:
    outputs:
      should_run: ${{ steps.logger_manifest.outputs.should_run }}