v0.9.6

Full Changelog: v0.9.5...v0.9.6

v0.9.5

Full Changelog: v0.9.4...v0.9.5

v0.9.4

Highlights\n\n- Add session/agent-scoped knowledge partitions with freeze and review-gated canonical merge.\n- Add pointer-first Context Packs with deduplication, sealed reuse, dependency binding, provenance protection, governance withholding, deterministic ordering, and byte budgets.\n- Add regression coverage for partition lifecycle, merge provenance, context reuse, tampering, conflicts, malformed inputs, and budget behavior.\n- Update Codex/Claude skill documentation and installation metadata.\n\n## Verification\n\n- 479 tests passed, 3 skipped.\n- Focused release checks passed: 11 context/partition tests, JSON validation, Python compilation, and diff checks.\n- Serialized byte counts are proxy metrics; host tokenizer measurement is still required for production token savings.

v0.9.3

Highlights

• Improved knowledge continuity, session checkpointing, and task reuse.
• Added evidence-resolution flow for missing knowledge and clearer project-bound retrieval.
• Added plain-language documentation and Archify workflow scenarios.
• Added Claude Code compatibility bridge and expanded validation coverage.

Verification

• 133 focused packaging, continuity, checkpoint, and knowledge-engineering tests passed.
• Archify workflow validation: 9/9, 0 errors, 0 warnings.
• Browser visual verification remains unverified because Chrome exited with SIGABRT.

v0.9.2

Highlights

• Hardened evidence selection so same-source, same-line claims with different labels are preserved.
• Added bounded structured-identifier prioritization for code and knowledge lookups.
• Added selector regression tests, discriminative benchmarks, balanced held-out evaluation, and locator fixtures.
• Kept group-v2 opt-in; legacy remains the default until the operational size gate is met.

Verification

• Skill package validation: passed
• Test suite: 416 passed, 3 skipped
• Held-out final replay: 8/8 locator checks for both selectors across 5 rounds
• git diff --check: passed

v0.9.1

Map-refresh lifecycle hooks now skip projects that have not activated Context Graph, avoiding repeated Claude hook errors caused by missing project configuration.

• SessionStart, SubagentStop and PostCompact return exit 0 with status=skipped and reason=project_not_activated when no project config exists.
• Existing, malformed, symlinked and foreign configurations still reach the project-bound validator; explicit retrieval/build commands retain their fail-closed behavior.
• Session recording remains a no-op without capture activation. No configuration, memory or map is created by the inactive hook.
• Adds regression tests and clarifies the activation contract.

Validation: 75 focused hook/plugin/session tests passed; full local project suite 370 passed, 3 skipped; packaging/version, knowledge-cycle, workspace and diff checks passed.

v0.9.0

Supporting-evidence audit (audit_memory_support.py), explicit session preflight in the knowledge-engineering skill, isolated-mode (python -I) sibling import fix for build_task_continuity.py, release version-alignment gate. Commit 4f0f005.

v0.8.0

Project-bound knowledge memory and continuity integration. Includes synchronized Codex and Claude Code runtime scripts, fail-closed project binding, session continuity records, packaging parity tests, and Claude Code host verification.

v0.7.0

Add multilingual code knowledge indexing. Python keeps commit-bound AST extraction; registered non-Python languages use low-confidence lexical candidates with Git-pinned locators, explicit parser/confidence metadata, diagnostics, and no unsupported structural relationship claims. Unknown extensions can be indexed explicitly through the lexical fallback.

v0.6.1

Package the Context Graph runtime inside the Codex skill so GitHub-installed skills can query and rebuild project-local knowledge maps. Update plog-post hooks to use its project-bound runtime.