anti-legacy-develop-plugin

name	anti-legacy:develop-plugin
description	Guide the agent in modifying the plugin's code, templates, or skills to address learnings and gaps found in UAT/semantic validation. Use when: "update the plugin", "fix parsing bug", "modify translation rule", "develop the plugin", "self-correct code translator based on learning".

anti-legacy:develop-plugin

This skill guides the agent in modifying the anti-legacy plugin itself — its code, templates, and skills — to address learnings and gaps surfaced in UAT verdicts, semantic validation findings, or production modernization runs.

Workflow

Step 1: Map the learning to the plugin component

Identify which plugin component requires modification:

• Code-graph / indexing bugs (missing dependencies, incorrect entities, missing data models, wrong SymbolId resolution, annotation not persisting):
  • The legacy code graph is now produced by wicked-estate (indexed by survey), not a local parser. There is no graph_builder.py. Structural gaps in the graph are wicked-estate concerns; surface them via the engine, not by editing a parser.
  • Modify scripts/wicked_estate.py (the integration helper — binary resolution, index/query/blast_radius/stats/cross_graph, and the resolve_symbol_id/annotate/read_semantics/by_requirement annotation path).
• Extraction / coverage errors (a behavior-bearing node left unaccounted, a rule annotated with wrong confidence/provenance, ring expansion not gathering enough context, coverage report miscounting the denominator):
  • Modify skills/extraction/SKILL.md (the adaptive ring-expansion crawl recipe that replaces the old graph-translator enrich flow).
  • Modify scripts/coverage.py (the resolved-or-flagged metric + behavior-bearing denominator) and the coverage.behavior_kinds / crawl.* keys in config.json.
• Requirements-scaffold drafting (the structural→functional requirements graph draft is wrong):
  • Modify scripts/graph_normalizer.py (drafts the requirements scaffold from the annotated graph; unchanged in WF1, no longer the code-graph parser).
• Translation / Code Swarm errors (incorrect target class structure, wrong types, missing imports):
  • Modify skills/swarm/SKILL.md (developer swarm prompt guidelines)
• Test Strategy / Contract mismatch (invalid test scenarios, missing rounding verification):
  • Modify skills/test-strategy/SKILL.md or scripts/test_runner.py
• Architectural / Compliance gaps (line-by-line translation, missing safety checks):
  • Modify skills/setup/references/anti_patterns.md or skills/blueprint/references/nfrs.md

Step 3: Write a regression test

Before writing code fixes, add a test case that captures the bug or gap:

• For helper / annotation bugs (binary resolution, SymbolId resolution, the silent-no-op guard, stats-digest determinism): add an assertion to tests/test_wicked_estate.py.
• For extraction / coverage bugs (denominator, resolved-or-flagged state, the unaccounted list): add an assertion to tests/test_coverage.py.
• For translation/semantic validation: add an assertion to tests/test_demo_pipeline.py or write a dedicated test file under tests/.

Step 4: Implement the fix

Modify the target files in the plugin codebase. Maintain documentation integrity and preserve unrelated docstrings or comments.

Step 5: Verify the fix

Run the specific test and then the full test suite to guarantee compile and runtime integrity:

# Run the specific test file
python3 -m unittest tests/test_my_fix.py -v

# Run the full test suite
python3 -m unittest discover -s tests -v

Halt and correct if any test fails.

Step 6: Commit the changes

Commit the fix to the git repository:

git add scripts/ skills/ templates/ tests/
git commit -m "plugin: fixed {bug_desc} based on {finding_source} finding {finding_id}"