Skip to content

LOG.md

Latest commit

 

History

History
82 lines (65 loc) · 4.51 KB

LOG.md

File metadata and controls

82 lines (65 loc) · 4.51 KB
Raw

2026-07-30

Changes

  • Added regression coverage for completion-order stability, module-ID collisions, cache provenance, atomic publication, writer locking, explicit roots/output, additional literal TypeScript imports, quoted paths, malformed maps, unsafe patterns, and invalid depths.
  • Added exact Python dependency pins and explicit builder/query path contracts.

Fixes

  • Resolved CALLS_API providers through consumer imports and failed closed when an unqualified provider name remained ambiguous.
  • Sorted parsed results and rejected every colliding normalized module ID before cache or map publication.
  • Added cache/map schema validation, builder fingerprints, atomic replacement, and a per-project non-blocking writer lock.
  • Replaced user-controlled regular expressions with a bounded linear pattern grammar and surfaced numeric, source-read, and structural map errors.
  • Encoded unsafe paths as quoted definitions and recognized TypeScript import = require() and literal template imports.

Motivation

  • GitHub issue #1 reproduced silent semantic changes, discarded modules, stale cache trust, unsafe concurrent writes, ambiguous targeting, and unbounded query input. These changes make correctness deterministic and failures local, explicit, and non-destructive.

2026-07-14

Changes

  • Added focused regression coverage for TSX imports, re-exports, literal runtime imports, project-scoped aliases, exact aliases, and package self-imports.

Fixes

  • Resolved TypeScript and TSX dependencies through the same module resolver.
  • Kept duplicate path aliases within their owning tsconfig.json boundary and supported exact aliases plus multiple configured targets.
  • Added import edges for export ... from, literal import(), and literal require() forms.
  • Resolved exact package self-imports to the nearest owning package source entrypoint without shadowing installed dependencies.

Motivation

  • An author-run external monorepo compiler oracle measured 99.899% local dependency recall and 99.832% precision. The oracle and dataset are not committed here, so this is provenance for the change rather than a reproducible repository benchmark.

2026-07-13

Changes

  • Added focused regression evals for cache validity, Python and TypeScript import impact, LSP batch progress, source-root resolution, layer classification, deterministic output, language gating, and infrastructure diagnostics.
  • Made the existing belief-search tests load the checked-out repository script instead of a machine-specific installed path.

Fixes

  • Invalidated incremental cache entries by complete SHA-256 content hash instead of mtime and partial-file hashing.
  • Restored source-aware queries by resolving source paths from the directory containing the belief map.
  • Prevented reference-only LSP batches from looping without progress.
  • Included nested and parent-relative Python imports in dependency edges.
  • Resolved TypeScript ESM .js import specifiers to their checked-in .ts and .tsx sources.
  • Stopped backend modules named as stores from being classified as UI unless their path has a frontend boundary.
  • Removed generated timestamps from the semantic map and made empty infrastructure output name its supported formats.
  • Canonicalized the scan root once so macOS path aliases do not leak into module IDs.

Motivation

  • The audit findings were reproducible on isolated fixtures and a large polyrepo workspace; these changes address the smallest correctness blockers while leaving unsupported languages and infrastructure formats explicit.

2026-07-12

Changes

  • Removed the obsolete .agents infrastructure-topology design notes and their README tree entry.

Motivation

  • The topology builder is implemented in the source, so retaining its original design notes duplicated completed work and left an unused .agents directory.

2026-06-19

Changes

  • Updated the codespaces skill workflow so agents are directed to query the belief map before reading source and to use quick as a first-pass analyze shortcut.
  • Changed belief_search.py quick to run full analyze output instead of boundary-only output, and made it resolve legacy path-map aliases found in raw belief-map hits.
  • Added focused unit coverage for quick analyze output, alias resolution from raw hits, boundary files-only output, and no-match suggestions.

Motivation

  • Session-log review showed that agents usually used search and analyze after activation, but often queried too late and did not discover shortcut or boundary commands.