Releases: AgentToolkit/altk-evolve
v1.0.6
v1.0.6 (2026-04-03)
Bug Fixes
-
Add optional implementation_steps to Tip model and prompt (#124,
d373e7e) -
Clarify task status context in tip generation prompt (#124,
d373e7e) -
Completely remove all Roo references from install.sh (#130,
8e6c76d) -
Normalize implementation_steps to list[str] in combine_cluster (#124,
d373e7e) -
Remove Roo integration tests and fix install.sh 'all' platform (#130,
8e6c76d) -
Resolve CI failures for formatting and type-checking (#91,
29dba09) -
bob: Clarify entity count rule in learn skill (#122,
824e4d9) -
claude: Move marketplace.json back to .claude-plugin directory (#115,
18ccaa9) -
claude: Move marketplace.json to .claude-plugin directory (#107,
a54a976) -
claude: Rename plugin to evolve-lite in marketplace.json (#116,
2cae903) -
entity-io: Prevent concurrent write collisions on same slug (#91,
29dba09) -
entity-io: Reject unsafe entity type values before joining paths (#91,
29dba09) -
entity-io: Use start-of-line match for rationale header parsing (#91,
29dba09) -
evolve-lite: Format entity_io.py to satisfy ruff check (#134,
43c5951) -
evolve-lite: Guard against non-string entity type values (#122,
824e4d9) -
evolve-lite: Guard Stop hook against recursion and clean up noise (#134,
43c5951) -
evolve-lite: Tighten learn skill to only extract high-signal guidelines (#122,
824e4d9) -
justfile: Correct volume mount path in sandbox-prompt (#133,
6e2c787) -
platform-integrations: Add missing entity_io lib to Bob (#105,
a3be71d) -
platform-integrations: Add missing entity_io lib to Bob integration (#105,
a3be71d) -
platform-integrations: Clean up Bob evolve-lite skills (#109,
889c8f1) -
platform-integrations: Clean up trailing whitespace and add actionable recall instructions (#109,
889c8f1) -
platform-integrations: Handle SameFileError race in _safe_copy2 (#104,
d88baf1) -
platform-integrations: Improve YAML parsing and error handling in install script (#100,
6bfdbbb) -
platform-integrations: Resolve Step 0 vs Step 1 ordering conflict (#109,
889c8f1) -
platform-integrations: Skip copy when src and dst are same file (#104,
d88baf1) -
postgres: Derive vector schema from embedding model (#135,
a6b7b33) -
postgres: Ensure pgvector extension before vector registration (#135,
a6b7b33) -
tests: Add postgres dependencies to dev group for testing (#128,
f2bc745) -
tests: Resolve linting errors and force isolated filesystem config in unit test fixture (#128,
f2bc745)
Documentation
-
Fix Codex Lite Mode installation steps in AGENTS.md (#136,
427e99d) -
Standardize cli commands to use evolve entrypoint across all readmes (#128,
f2bc745) -
Sync documentation with supported platforms (add codex, remove roo) (#136,
427e99d) -
codex: Align evolve-lite skill naming and expand Codex skill guidance (#137,
f962b55) -
evolve-lite: Document Stop hook UX implications and opt-out mechanism (#134,
43c5951)
Features
v1.0.5
v1.0.4
v1.0.3
v1.0.3 (2026-03-12)
Bug Fixes
-
save-trajectory: Address code review findings (#89,
6e6438b) -
save-trajectory: Make log() best-effort so debug logging never crashes the script (#89,
6e6438b) -
save-trajectory: Restrict file permissions on trajectories dir and output files (#89,
6e6438b) -
save-trajectory: Use EXIT trap for temp file cleanup to handle script failures (#89,
6e6438b) -
save-trajectory: Validate trajectory type and broaden input error handling (#89,
6e6438b)
Documentation
Features
-
Add save-trajectory skill to export conversation as OpenAI chat format (#89,
6e6438b) -
backend: Refactor entity update with template method pattern and upgrade dependencies (#86,
50efcd2)
Detailed Changes: v1.0.2...v1.0.3
v1.0.2
v1.0.1
v1.0.0
v1.0.0 (2026-03-04)
Bug Fixes
-
Add Pydantic validation error handling for LLM tip generation and validate trajectory data input. (#56,
d2a4d0a) -
Address CodeRabbit review feedback on tip clustering (#60,
8ea2051) -
Clean up stale variable name and deduplicate entity file locations (#67,
9dcdc54) -
Correct typo in EXIF example run instructions (#80,
0b3a8e8) -
Enhance LLM tip generation robustness by skipping empty assistant messages and handling malformed/empty responses, and add validation for trajectory data. (#56,
d2a4d0a) -
Guard against empty tips list in MCP save_trajectory (#60,
8ea2051) -
Guard against empty tips list in MCP save_trajectory (#58,
ce1ead3) -
Harden milvus filters and fact extraction input handling (
36dee90) -
Narrow created_at type handling for mypy (
36dee90) -
Prevent shell injection in sandbox-prompt via env variable (#80,
0b3a8e8) -
Resolve filesystem delete bug and update MCP namespace logic (#68,
16aeb78) -
Run sandbox container as non-root user and harden installer (#75,
647aea0)
Documentation
-
Add tie-breaker rule for entity count cap in learn skill (#69,
9721a05) -
Fix sandbox README paths to work from repo root (#75,
647aea0) -
Move LiteLLM proxy details to configuration guide (
36dee90) -
agents: Add conventional commits format guidance for python-semantic-release (#82,
2c715dc)
Features
-
Add error-prevention focus to learn skill and skip known-failed approaches in recovery (#69,
9721a05) -
Add policy support and restore mcp backward compatibility (#68,
16aeb78) -
Add sandbox demo tooling and EXIF extraction example (#80,
0b3a8e8) -
Add sandbox for running Claude Code in Docker (#75,
647aea0) -
Cluster tips by task description cosine similarity (#60,
8ea2051) -
Combine tips within clusters via LLM consolidation (#60,
8ea2051) -
Move entity storage to .kaizen/ and add Kaizen Lite guide (#67,
9dcdc54) -
Persist task description in tip entity metadata (#60,
8ea2051) -
Persist task description in tip entity metadata (#58,
ce1ead3) -
config: Support LiteLLM proxy env mapping and document model precedence (
36dee90)
Testing
Detailed Changes: v0.2.1...v1.0.0