Consolidate dead duplicate `resolve_*` family in `config/feature_configs.py` onto canonical `param_resolver`

[MervinPraison]

Summary

praisonaiagents/config/feature_configs.py carries a second, parallel implementation of the parameter-precedence resolvers that the SDK already provides in praisonaiagents/config/param_resolver.py (which self-documents as "the SINGLE, DRY resolver"). The feature_configs versions have no production callers — every public config.resolve_* name routes to param_resolver — and are kept alive only by one precedence-ladder test. This is duplicate resolver logic that can be consolidated without dropping any option or behaviour.

Current behaviour

Dead duplicate resolvers in config/feature_configs.py (none referenced by production code):

• resolve_memory (1274), resolve_knowledge (1346), resolve_planning (1400), resolve_reflection (1413), resolve_guardrails (1426), resolve_web (1441), resolve_caching (1455), resolve_autonomy (1468)
• resolve_tool_search (1481) — zero references anywhere in the codebase (not even tests).

Routing evidence:

• config/__init__.py maps every resolve_* name to "param_resolver" (lines 136–148). None map to feature_configs.
• Top-level praisonaiagents/__init__.py maps resolve_memory/knowledge/web/planning/reflection/autonomy/caching/guardrails to praisonaiagents.config.param_resolver (lines 424–437).
• agent/agent.py:1430 imports resolve_guardrails from param_resolver (not feature_configs).
• The only importers of the feature_configs resolver functions are tests (e.g. tests/unit/config/test_precedence_ladder.py). No production path imports them.

Not dead — keep: resolve_tools (feature_configs.py:1496) is used by agent.py:2070 and has no param_resolver equivalent. The *Config dataclasses in feature_configs.py are widely imported and must stay.

Why it matters

Maintenance: the precedence ladder (Instance > Config > Array > Dict > String > Bool > Default) is implemented twice. A fix or new option added to param_resolver silently diverges from the shadow copy in feature_configs. Note the two families also have different signatures (feature_configs: single-arg resolve_memory(value); param_resolver: resolve_memory(value, config_class)), so the duplication is also a subtle inconsistency.

Category

Duplicate / Dead code

Capability preserved

• param_resolver remains the canonical resolver — every public config.resolve_* name keeps working unchanged.
• All precedence rules and accepted input forms (Instance/Config/Array/Dict/String/Bool/Default) are retained.
• resolve_tools and resolve_tool_search decisions: resolve_tools kept as-is (used in production); resolve_tool_search retained or shimmed for backward-compat even though unreferenced — no public name removed.
• All *Config dataclasses untouched.

Proposed approach

Make param_resolver canonical and reduce the dead feature_configs resolvers to delegating wrappers (or repoint the single precedence-ladder test at param_resolver and drop the dead bodies). Preserve every public resolve_* name and the *Config dataclasses.

Resolution sketch

# Before: config/feature_configs.py — full second implementation
def resolve_memory(value: MemoryParam) -> Optional[MemoryConfig]:
    # ...precedence ladder re-implemented...

# After: thin delegation to the canonical resolver (keeps the name + signature)
def resolve_memory(value: MemoryParam) -> Optional[MemoryConfig]:
    from .param_resolver import resolve_memory as _resolve
    return _resolve(value, MemoryConfig)

Severity

Low

Validation

• Traced: resolve_memory/knowledge/planning/reflection/guardrails/web/caching/autonomy in feature_configs.py have zero production references (only param_resolver/__init__ routing + one test).
• resolve_tool_search has zero references anywhere.
• Confirmed resolve_tools IS used (agent.py:2070) — explicitly kept.
• No public API name removed; behaviour and all options preserved.

Keep unchanged

• param_resolver.py (canonical resolver) and all its precedence logic.
• resolve_tools in feature_configs.py.
• All *Config dataclasses (MemoryConfig, WebConfig, OutputConfig, MultiAgent*Config, etc.) and the public config.resolve_* names.

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job • claude/issue-2137-20260621-1710 • [Create PR ➔](https://github.com/MervinPraison/PraisonAI/compare/main...claude/issue-2137-20260621-1710?quick_pull=1&title=Issue%20%232137%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%232137%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

Todo List

• Read AGENTS.md to understand architecture rules
• Analyze duplicate resolver functions in feature_configs.py
• Verify usage in param_resolver.py
• Identify which resolvers are truly dead code
• Replace duplicates with delegating wrappers
• Ensure resolve_tools is kept unchanged
• Test the changes
• Commit and push the fix

Completed! ✅

Successfully consolidated duplicate resolver functions in feature_configs.py by replacing them with thin delegation wrappers to the canonical param_resolver.py versions.

Changes made:

• Replaced 8 duplicate resolver implementations with delegation wrappers
• Kept resolve_tool_search with simple implementation (zero references but preserved for backward compatibility)
• Preserved resolve_tools unchanged (used in production)
• All public API names and signatures maintained

Impact:

• Eliminated ~100 lines of duplicate code
• Single source of truth for precedence ladder logic
• Fully backward compatible with existing tests

Pull Request: #2140
|