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
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Added
Clarify tool -- agent-initiated clarify() asks the user a question mid-run and blocks until answered. Works in REPL, bot, daemon, and autonomous modes. Add { type: clarify } to your role's tools list.
Context budget guard -- PydanticAI history processor that truncates oversized parts and drops oldest message pairs when estimated tokens exceed the budget. Runs as a pre-request history processor and between autonomous/daemon iterations.
model.context_window field -- explicit context window override in spec.model; auto-detected per provider when omitted (200k Anthropic, 128k OpenAI/Gemini, etc.)
Dashboard trigger visibility panel -- agent detail page shows trigger configuration, fire counts, and next-run times
Dashboard toast notification system -- global toast store with auto-dismiss; load-error recovery across all pages
Dashboard skills management -- browse, search, create, and edit skills from the dashboard. Path-based skill discovery, scope badges, requirement indicators, config panel, and editor tab. Full CRUD API via /api/skills/ endpoints.
Skill service layer -- SkillService with path-based discovery, SKILL.md parsing, validation, and CRUD operations (initrunner/services/skill_service.py)
Agent skills sidebar -- agent config panel shows attached skills with scope and requirement badges
Trigger stats API -- GET /api/agents/{name}/trigger-stats endpoint with fire counts and schedule info
Context-aware-scraper example -- autonomous web scraper demonstrating the context budget guard with history summarization
Clarify-researcher example -- research assistant demonstrating human-in-the-loop clarify tool before acting
Removed
kind: Pipeline orchestration -- removed in favor of Team for one-shot multi-agent workflows and Compose for long-running services. For complex DAG workflows, use pydantic-graph directly. Running a Pipeline YAML now shows a migration message.
--var flag from initrunner run (was Pipeline-only)
Pipeline executor, schema, loader, and tests (initrunner/pipeline/)
Changed
Compose scaffold pattern renamed -- --pattern pipeline is now --pattern chain. Default for initrunner compose new updated accordingly.
Dashboard compose UI no longer labels runs or deletions as "pipeline"