chore: PDD sync for #879

[prompt-driven-github]

Summary

Adds the compression_benchmark_python.prompt that generates a benchmark orchestrator comparing five compression strategies across frozen fixtures, plus the contract_loss_regression fixture set that proves contract-symbol preservation is required. Updates docs/ci.md with the new benchmark gate documentation.

Closes #879

Changes Made

Prompts Modified

• pdd/prompts/compression_benchmark_python.prompt — new prompt generating pdd/compression_benchmark.py, a benchmark orchestrator with five strategy callables (full_tests, ast_tests, ast_contracts, full_fewshot, compressed_fewshot), CompressionBenchmarkResult dataclass, check_compression_benchmark() gate enforcement, and Markdown+JSON report output

Fixtures Added

• tests/fixtures/compression_benchmark/contract_loss_regression/source.py — source module with contract symbols
• tests/fixtures/compression_benchmark/contract_loss_regression/test_file.py — pytest file for the fixture scenario
• tests/fixtures/compression_benchmark/contract_loss_regression/contract_seeds.json — required contract symbol seeds
• tests/fixtures/compression_benchmark/contract_loss_regression/expected_output.py — canonical expected output
• tests/fixtures/compression_benchmark/contract_loss_regression/expected_output_*.py — per-strategy expected outputs
• tests/fixtures/compression_benchmark/contract_loss_regression/expected_results.json — per-strategy pass/token/churn baselines showing ast_tests and full_fewshot fail without contract-source preservation

Documentation Updated

• docs/ci.md — new "Compression Benchmark Gate" section documenting five strategies, five metrics, three hard-failure conditions, fixture schema, and CI placement guidance

Architecture Updated

• architecture.json — registers compression_benchmark module with its dependencies and interface

Review Checklist

• Prompt syntax is valid
• PDD conventions followed
• Documentation is up to date
• Fixture baseline values reflect realistic token counts and churn thresholds
• expected_results.json correctly shows contract-loss failure for ast_tests / full_fewshot strategies

Next Steps After Merge

1. Regenerate code from modified prompts in dependency order:

   pdd sync compression_benchmark

2. Run tests to verify functionality:

   pytest -vv tests/test_compression_benchmark.py

3. Deploy if applicable

---

Created by pdd change workflow

[prompt-driven-github]

Step 7/8: Review Loop Final Report

PR: #1396
Issue: #879
issue_aligned: false
active-reviewer: codex
reviewer-status: codex=findings claude=fixer fresh-final=missing
fresh-final-review: missing
verified-head-sha: none
remote-pr-head-sha: none
max-rounds-reached: false
max-cost-reached: false
max-duration-reached: false

Summary

Fixer claude could not address codex's findings.

Per-Reviewer Status

Reviewer	Status
codex	findings
claude	fixer
fresh-final	missing

Findings

Severity	Status	Location	Finding	Required fix	Reviewer
critical	open	pdd/compression_benchmark.py:44	The benchmark gate cannot run on the PR's own frozen fixture. This leaves the user workflow incomplete: CI/users get a crash from the AST test strategy instead of the required comparison and gate result.	Change the AST-test strategy to use selectors that are valid for test files, or handle test slicing separately from source contract seeds, and add an unmocked integration test that runs the checked-in fixture through run_benchmark and check_compression_benchmark successfully.	codex
medium	open	pdd/compression_benchmark.py:178	The gate does not prove compression reduces context size, which is the core issue goal. Token counts are reported but not enforced against the frozen baselines or the full-context strategy.	Add a token regression rule to the prompt and implementation, for example failing when res.input_tokens > baseline['max_input_tokens'] and/or when compressed strategies are not smaller than their full-context comparator, and add tests that demonstrate the gate fails on token-budget regressions.	codex
medium	open	pdd/prompts/compression_benchmark_python.prompt:15	The prompt source of truth is missing the gate_python.prompt dependency even though the module depends on gate types. Future architecture regeneration can drop or disagree with the current architecture dependency graph.	Add <pdd-dependency>gate_python.prompt</pdd-dependency> to the compression benchmark prompt metadata and keep the generated architecture entry synchronized.	codex

Fixer Rationale

• pdd/compression_benchmark.py:44: The benchmark gate cannot run on the PR's own frozen fixture. This leaves the user workflow incomplete: CI/users get a crash from the AST test strategy instead of the required comparison and gate result. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":429,"duration_ms":651,"duration_api_ms":0,"num_turns":1,"result":"You've hit your limit · resets 9:30pm (UTC)","stop_reason":"stop_sequence","session_id":"34146e3d-7e24-41d9-8a12-4c1accb1e22e","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"server_tool_use":{"web_search_requests":0,"web_fetch_reques'; verification=unverified)
• pdd/compression_benchmark.py:178: The gate does not prove compression reduces context size, which is the core issue goal. Token counts are reported but not enforced against the frozen baselines or the full-context strategy. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":429,"duration_ms":651,"duration_api_ms":0,"num_turns":1,"result":"You've hit your limit · resets 9:30pm (UTC)","stop_reason":"stop_sequence","session_id":"34146e3d-7e24-41d9-8a12-4c1accb1e22e","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"server_tool_use":{"web_search_requests":0,"web_fetch_reques'; verification=unverified)
• pdd/prompts/compression_benchmark_python.prompt:15: The prompt source of truth is missing the gate_python.prompt dependency even though the module depends on gate types. Future architecture regeneration can drop or disagree with the current architecture dependency graph. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":429,"duration_ms":651,"duration_api_ms":0,"num_turns":1,"result":"You've hit your limit · resets 9:30pm (UTC)","stop_reason":"stop_sequence","session_id":"34146e3d-7e24-41d9-8a12-4c1accb1e22e","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"server_tool_use":{"web_search_requests":0,"web_fetch_reques'; verification=unverified)

Fixes Attempted

• round=1 fixer=claude fixer_result=failed push_status=not_attempted local_sha=none pushed_sha=none changed_files=none verification=unverified summary=All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":429,"duration_ms":651,"duration_api_ms":0,"num_turns":1,"result":"You've hit your limit · resets 9:30pm (UTC)","stop_reason":"stop_sequence","session_id":"34146e3d-7e24-41d9-8a12-4c1accb1e22e","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"server_tool_use":{"web_search_requests":0,"web_fetch_reques

[prompt-driven-github]

Step 7/8: Review Loop Final Report

PR: #1396
Issue: #879
issue_aligned: false
active-reviewer: codex
reviewer-status: codex=findings claude=fixer fresh-final=missing
fresh-final-review: missing
verified-head-sha: none
remote-pr-head-sha: none
max-rounds-reached: false
max-cost-reached: false
max-duration-reached: false

Summary

Fixer claude could not address codex's findings.

Per-Reviewer Status

Reviewer	Status
codex	findings
claude	fixer
fresh-final	missing

Findings

Severity	Status	Location	Finding	Required fix	Reviewer
critical	open	pdd/compression_benchmark.py:44	The benchmark gate cannot run on the PR's own frozen fixture. A real CI/user workflow gets a crash while building the ast_tests or compressed_fewshot context instead of receiving the required five-strategy comparison and GateResult.	Change the AST-test strategy to use selectors that are valid for pytest files, or handle test slicing separately from source contract seeds, then add an unmocked integration test that runs the checked-in fixture through run_benchmark and check_compression_benchmark successfully. The current tests mock ContentSelector and ApiContractSlicer around the real fixture, so they do not prove the shipped workflow works.	codex
medium	open	pdd/compression_benchmark.py:178	The gate reports token counts but does not prove compression reduces context size, which is the core issue goal and is also promised by docs/ci.md. Token regressions or compressed strategies that are larger than their full-context comparator can pass the benchmark gate.	Add a token regression rule to the prompt and implementation, for example failing when res.input_tokens > baseline['max_input_tokens'] and/or when compressed strategies are not smaller than their full-context comparator. Add tests showing the gate fails on token-budget regressions.	codex
medium	open	pdd/prompts/compression_benchmark_python.prompt:15	The prompt source of truth is missing the gate dependency, so repository architecture/include validation fails and future regeneration can drift from the checked-in architecture entry.	Add <pdd-dependency>gate_python.prompt</pdd-dependency> to pdd/prompts/compression_benchmark_python.prompt and keep the generated architecture metadata synchronized.	codex

Fixer Rationale

• pdd/compression_benchmark.py:44: The benchmark gate cannot run on the PR's own frozen fixture. A real CI/user workflow gets a crash while building the ast_tests or compressed_fewshot context instead of receiving the required five-strategy comparison and GateResult. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":403,"duration_ms":454,"duration_api_ms":0,"num_turns":1,"result":"Your organization does not have access to Claude. Please login again or contact your administrator.","stop_reason":"stop_sequence","session_id":"f6df72d4-5d41-4388-abb9-acf0f025bc10","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"ser'; verification=unverified)
• pdd/compression_benchmark.py:178: The gate reports token counts but does not prove compression reduces context size, which is the core issue goal and is also promised by docs/ci.md. Token regressions or compressed strategies that are larger than their full-context comparator can pass the benchmark gate. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":403,"duration_ms":454,"duration_api_ms":0,"num_turns":1,"result":"Your organization does not have access to Claude. Please login again or contact your administrator.","stop_reason":"stop_sequence","session_id":"f6df72d4-5d41-4388-abb9-acf0f025bc10","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"ser'; verification=unverified)
• pdd/prompts/compression_benchmark_python.prompt:15: The prompt source of truth is missing the gate dependency, so repository architecture/include validation fails and future regeneration can drift from the checked-in architecture entry. (fixer=claude fixer_summary='All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":403,"duration_ms":454,"duration_api_ms":0,"num_turns":1,"result":"Your organization does not have access to Claude. Please login again or contact your administrator.","stop_reason":"stop_sequence","session_id":"f6df72d4-5d41-4388-abb9-acf0f025bc10","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"ser'; verification=unverified)

Fixes Attempted

• round=1 fixer=claude fixer_result=failed push_status=not_attempted local_sha=none pushed_sha=none changed_files=none verification=unverified summary=All agent providers failed: anthropic: Exit code 1: {"type":"result","subtype":"success","is_error":true,"api_error_status":403,"duration_ms":454,"duration_api_ms":0,"num_turns":1,"result":"Your organization does not have access to Claude. Please login again or contact your administrator.","stop_reason":"stop_sequence","session_id":"f6df72d4-5d41-4388-abb9-acf0f025bc10","total_cost_usd":0,"usage":{"input_tokens":0,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":0,"ser