You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Added xtforge.core.code.audit (audit_workspace), detecting nested-loop duplicate aggregation without deduplication, asymmetric/off-by-one boundary comparisons (< vs <=), stale module caches where mutators omit invalidation, integer floor-division truncation in allocation helpers, and mutable defaults, with full line-numbered function_source and reader_source evidence.
Cross-Module Active Dispatch Tracer (xt-forge trace / code trace):
Added xtforge.core.code.trace (trace_symbol), resolving runtime execution hops across package __init__.py re-exports (from ... import ... as ...), duplicate symbol names, and JSON/YAML/TOML/INI/ENV/Python configuration dispatch tables (active_implementation, active_path, disputed_edges, and inactive_candidates with line-numbered function excerpts).
Added top-level aliases xt-forge callers, xt-forge audit, xt-forge trace, xt-forge diff, and xt-forge context, plus first-class --meta MCP tools code_callers, code_audit, code_trace, and re_diff with zero-dash argument safety and evidence_complete flags in both Go (xtforge-hub) and Python (xtforge/adapters/mcp_server.py).
24 XTF-* Workflow Skills:
Added 6 new workflow skills (XTF-CodeAudit, XTF-DispatchTrace, XTF-BinaryDiff, XTF-LSPNavigate, XTF-SwarmRE, XTF-SessionRecovery), expanding the catalog to 24 skills across Simple (8), Complex (8), and Advanced (8) tiers.
Fixed
6-Tier Live Benchmark (scripts/bench_three_tier.py):
Extended the benchmark suite to 6 tiers (tier1_find, tier2_fix, tier3_recover, tier4_defects, tier5_trace, tier6_regression_src, tier6_regression_bin) with live A/B pairing against raw REA (rea-agents, 229 tools) and fixed grade_tier1 so call-site grep lines inside fenced code blocks are not misclassified as definition bindings.