perf(lockfile-stats): enforce single-script cached analysis to cut token/turn cost

[Copilot]

✨ Enhancement

The lockfile-stats workflow prompt was driving high-cost runs (11–15M tokens, 51 bash turns) by repeatedly reading raw lock YAML across hundreds of files. This change reshapes the prompt into a bounded, cache-first analysis flow targeting ≤1M tokens and ≤5 bash turns using compact JSON summaries.

• What does this improve?

  • Forces a single first-turn analyzer execution that parses all lockfiles once and emits one compact summary JSON.
  • Shifts reasoning from raw YAML to /tmp/gh-aw/agent/lockfile-stats-summary.json only.
  • Adds explicit run budgets and stop behavior (token/turn ceilings + noop fallback).

• Why is this valuable?

  • Eliminates repeated raw lockfile reads from the model context.
  • Reduces turn churn and token amplification from iterative shell/YAML inspection.
  • Makes runtime behavior deterministic via bounded output and explicit truncation priorities.

• Implementation approach:

  • Updated .github/workflows/lockfile-stats.md to define:
    • strict performance contract (effective tokens, bash-turn counting, budget stop condition),
    • cache-memory script reuse/versioning (lockfile_stats_v1.py, version bump rules),
    • compact summary constraints (size target + deterministic truncation order),
    • JSON-only downstream analysis and optional historical delta flow.

### 1) First turn: run one command that caches + executes the analyzer
...
4. Produces `/tmp/gh-aw/agent/lockfile-stats-summary.json` (compact, target ≤50KB; if larger, reduce examples before writing).
5. If the prompt version is bumped (for example to `lockfile_stats_v2.py`), do not reuse older script versions; use the version referenced in this prompt.

[Copilot]

Pull request overview

Refactors the lockfile-stats workflow prompt to enforce a bounded, cache-first analysis flow that avoids repeated raw lockfile reads and constrains bash turns/tokens.

Changes:

• Adds a strict performance contract (effective token + bash turn ceilings and a stop/noop fallback).
• Requires a single first-turn analyzer script run that writes a compact JSON summary for downstream reasoning.
• Defines cache-memory/script versioning and optional historical delta reporting from cached summaries.

Show a summary per file

File	Description
.github/workflows/lockfile-stats.md	Replaces the open-ended, multi-phase instructions with a single-script cached JSON analysis flow + explicit runtime budgets.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 1/1 changed files
• Comments generated: 2