Automated AST Migration and Verification Pipeline Specification

Role: Staff Software Quality Engineer specializing in automated developer tooling, AST code transformations, and CI/CD verification chains.

Context

• Source Repository & Language: {{source_repository}}
• Target Migration Standard: {{target_framework_version}}
• Test Verification Harness: {{verification_test_suite}}
• Agent Router Architecture: {{llm_router_config}}
• Human Review Gate: {{human_review_gate}}
• Pull Request Policy: {{pull_request_policy}}

Task

Produce an Engineering Pipeline Specification for an autonomous multi-agent refactoring workflow chain that ingests {{source_repository}}, performs syntactically valid upgrades to {{target_framework_version}}, validates output against {{verification_test_suite}}, and adheres to {{pull_request_policy}}.

Method

1. Establish file discovery and partitioning logic to chunk the codebase within {{source_repository}} by dependency graph order.
2. Define prompt and tool-call contracts for the Analyzer Agent to identify deprecated APIs and incompatible syntax.
3. Define the Transformation Agent spec, detailing AST-aware patch generation targetting {{target_framework_version}}.
4. Design the Linter and Static Analysis Agent step that executes local compile checks before test running.
5. Specify the Test Runner Agent protocol that executes {{verification_test_suite}} and feeds stack traces back on failure.
6. Detail an automated self-healing loop permitting a maximum of 3 localized remediation cycles on test failure.
7. Define the boundary conditions where {{human_review_gate}} is triggered versus automated PR submission under {{pull_request_policy}}.

Constraints

• MUST NOT permit any transformed code to bypass {{verification_test_suite}} prior to PR generation.
• Generated code patches MUST maintain byte-level deterministic formatting matching repository lint rules.
• The self-healing loop MUST abort and trigger {{human_review_gate}} if the same test fails consecutively twice.
• All agent token budgets and execution timeouts must be bounded per file.

Output format

1. Multi-Agent Pipeline Topology (Workflow diagram, step dependencies, and tool execution boundaries)
2. Agent Node Specifications (Individual agent inputs, outputs, system prompt constraints, and allowed tool calls)
3. Self-Correction & Feedback Loop Protocol (Stack trace parsing, prompt injection format, and backoff limits)
4. Quality Gate & PR Lifecycle Policy (Criteria for mergeability, diff size limits, and {{human_review_gate}} escalation rules)

Self-review

• Ensure dependency-ordered processing prevents circular refactoring conflicts.
• Verify self-healing loop has a deterministic exit condition to prevent infinite LLM token spend.
• Check that {{target_framework_version}} and {{verification_test_suite}} constraints are strictly enforced in the quality gates.