Forge: preserve parallel action identity

[GolaraJ]

Summary

• emit every controller action with description equal to its stable actionId
• require that description to be passed unchanged to session_store_sql
• prohibit mapping parallel outcomes by ordinal or completion order
• add controller coverage for stable discovery and tool-action descriptions

Root cause

The latest run swapped two parallel outcomes: the first tool action timed out and the second succeeded, but Forge attempted to materialize the timed-out action and recorded the successful action as timed out. The materializer correctly found no result for the timed-out SQL.

Validation

• 48 Skill Forge tests pass
• Python scripts compile
• git diff --check passes

[Copilot]

Pull request overview

This PR addresses a parallel-execution identity bug by ensuring each controller-emitted action has a stable, non-ordinal identifier that can be carried through tool execution and outcome recording without relying on completion order.

Changes:

• Emit controller actions with description set to the stable actionId (for discovery, metadata, refs, files, and tool-calls actions).
• Document the requirement to pass description unchanged into session_store_sql, and to avoid mapping outcomes by ordinal/completion order.
• Add/extend controller tests to assert description == actionId for both discovery and parallel tool-call actions.

Show a summary per file

File	Description
plugins/repo-dreamer/skills/repository-skill-forge/scripts/extraction-controller.py	Adds description: action_id to all emitted action objects to preserve stable identity across parallel runs.
plugins/repo-dreamer/skills/repository-skill-forge/tests/test_extraction_controller.py	Adds assertions verifying action description matches actionId for discovery and tool-call actions.
plugins/repo-dreamer/skills/repository-skill-forge/SKILL.md	Documents the stable-identity requirement and warns against ordinal/completion-order outcome mapping.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 3/3 changed files
• Comments generated: 0
• Review effort level: Lite