5.2: Extract pipeline steps into standalone SkillCard files

[savitharaghunathan]

5.2: Extract pipeline steps into standalone SkillCard files

Part of Stream 5: Harness → AgentPlaybookRun alignment

Migration logic is currently embedded in Go packages (internal/plan, internal/execute, internal/verify, internal/fixloop) and custom recipe YAML files (harness/recipes/*.yaml). Per ADR-0001, each pipeline step should be a standalone SkillCard with a SKILL.md at /opt/skills/{name}/SKILL.md, following the skillimage.io/v1alpha1 format.

Tasks

• Create skills/plan/SKILL.md — reads repo, runs graphify, writes PLAN.md (no source changes)
• Create skills/execute/SKILL.md — reads PLAN.md, applies transformations file by file
• Create skills/verify/SKILL.md — runs build, parses errors, applies conservative fixes, iterates up to max iterations
• Add skill.yaml metadata for each skill (for future OCI packaging with skillimage)
• Include reference files alongside each skill (e.g. skills/execute/references/javaee-quarkus.md)
• Remove harness/skill-bundle/goose-migration/ embedded bundle
• Remove harness/recipes/*.yaml custom recipe format
• Remove recipe parser (internal/goose/recipe.go)

Acceptance Criteria

• Each pipeline step is a self-contained skill directory under skills/ with a SKILL.md and optional references/ subdirectory
• Skills follow the skillimage.io/v1alpha1 SkillCard format (YAML frontmatter with name, description, type)
• The harness Go code has no knowledge of pipeline steps — it discovers and loads whatever skill is at /opt/skills/*/SKILL.md
• Old recipe YAML format and embedded skill bundle are removed

References

• ADR-0001 §Skill mounting: one directory
• ADR-0001 §Skill and rule packaging via skillimage
• PR ✨ Harness init #33