- Rejected unknown helper CLI options and documented consistent
--key=valueparsing. - Strengthened sparse capture validation around
User requestplus substantive evidence. - Split detailed installation options into
INSTALLATION.mdwhile keeping README install quick. - Aligned plugin and agent metadata descriptions.
- Breaking: renamed repo-local capture state from
.ai/to.capture/, including default captures at.capture/raw/, config at.capture/config.yaml, and the active pointer at.capture/pointer.json. - Configured output roots receive capture files only; helper pointer and lock state stay in repo-local
.capture/. - Bumped raw capture schema version to
0.7. - Rewrote README as a human-facing installation and positioning page while keeping skill docs strict and agent-optimized.
- Added root Codex and Claude Code plugin manifests plus marketplace entries for plugin-based installation without duplicating skill files.
- Documented OpenCode and generic-agent direct skill install paths.
- Made helper invocation docs install-location-neutral so global and plugin installs resolve captures against the target repo.
- Shortened skill frontmatter to a discovery-focused description.
- Bumped raw capture schema version to
0.6. - Updated contract docs from v0.5.1 to v0.6 scope.
- Moved the source skill directory from
skill/knowledge-capturetoskills/knowledge-capture.
- Added MIT licensing.
- Added README positioning for wiki/LLM knowledge-base ingestion workflows.
- Separated marketing-facing README language from agent-optimized skill docs.
- Collapsed
changed_by_sourceinto compactchanged_byprovenance suffixes. - Clarified Codex session IDs as explicit runtime values, not assumed shell variables.
- Compacted skill docs while keeping raw capture schema version
0.5.
- Added configurable local output roots for vault/inbox placement via
capture.output_rootor--output-root. - Added stricter capture-quality guidance to keep durable context and drop receipt-like command noise.
- Added capture identity metadata:
agentandchanged_by. - Bumped raw capture schema version to
0.5.
- Clarified capture lifecycle: one capture per current agent session, scoped to its workflow.
- Guarded
--update-activeso it only updates when--agent-session-idmatches the active pointer. - Fixed
.ai/config.yamlparsing so one-level scalar sections accept any consistent positive indentation. - Compacted repo and skill instructions while preserving active-pointer, local-only, safety, and parser rules.
- Updated contract tests for session-boundary behavior, YAML indentation, compact wording, and schema version
0.4.
- Hardened raw capture behavior with lock-guarded active-pointer writes, stale-lock cleanup, and atomic pointer replacement.
- Broadened obvious secret-risk detection.
- Simplified raw capture sections to
User request,Outcome,Changes and evidence,Decisions and discoveries, andOpen questions and next steps. - Moved detailed runtime shape guidance into references and kept
SKILL.mdconcise. - Documented local-first
.ai/raw/posture and excluded sync, durable memory promotion, graph/vector databases, and Obsidian processing.