# agent-actions

> Declarative framework for orchestrating multi-model LLM pipelines with context engineering and quality gates.

- **URL**: https://www.freshcrate.ai/projects/agent-actions
- **Author**: Muizzkolapo
- **Category**: Testing
- **Latest version**: `v0.2.4` (2026-06-06)
- **License**: Apache-2.0
- **Source**: https://github.com/Muizzkolapo/agent-actions
- **Homepage**: http://docs.runagac.com/
- **Language**: Python
- **GitHub**: 7 stars, 1 forks
- **Registry**: github
- **Tags**: `ai-agents`, `anthropic`, `context-engineering-framework`, `llm`, `orchestration`, `prompt-engineering`, `prompt-engineering-tool`, `python`, `yaml`

## Description

Declarative framework for orchestrating multi-model LLM pipelines with context engineering and quality gates.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.2.4` | 2026-06-06 | High | ## Bug Fix * Observe directives on missing fields now skip the record instead of silently injecting None into prompts * Empty dict responses `[{}]` no longer bypass `_is_empty_output` check; reprompt exhaustion now routes through EXHAUSTED disposition instead of SUCCESS  ## Under the Hood * Release v0.2.3 — batch changie entries, bump version |
| `v0.2.3` | 2026-05-31 | High | ## v0.2.3 - 2026-05-31 ### Enhancement or New Feature * Add incremental checkpointing for online processing — interrupted runs resume from the last checkpoint instead of reprocessing all records * Add field visibility settings panel to HITL review UI — reviewers can configure which fields stay expanded vs collapsed per session ### Under the Hood * Release v0.2.2 — batch changie entries, bump version * Remove dead VendorRegistry class, SINGLE_RESPONSE_CLIENTS empty set, and stale dispatch_task co |
| `v0.2.2` | 2026-05-30 | High | ## v0.2.2 - 2026-05-30  ### Bug Fix * Add exponential backoff with jitter to batch retry loop to prevent retry storms when provider is overloaded * Stop silently graduating records on reprompt batch submission failure; transient and permanent errors now handled distinctly with proper metadata * Remove silent os.getenv API key fallbacks in batch client factory — batch clients now raise ConfigurationError when no key is configured instead of silently trying hardcoded vendor env vars * Log recovery |
| `v0.2.0` | 2026-05-23 | High | ## v0.2.0 - 2026-05-23 ### Breaking Change * Removed cross-workflow chaining: upstream: config key, --upstream/--downstream CLI flags, WorkflowOrchestrator, and VirtualAction injection. Workflows are self-contained. Users with upstream: in config will see an unknown-key warning.  ### Enhancement or New Feature * v0.2.0 Quality Hardening — 25 fixes from comprehensive codebase audit: - Storage errors now propagate instead of silently returning COMPLETED - Schema migration uses ALTER TABLE instead |
| `v0.1.17` | 2026-05-16 | High | ## What's new in v0.1.17  **40 commits since v0.1.16** — major reliability, observability, and UI improvements.  ### Highlights  **Pipeline reliability** - Record-level error isolation and selective retry (#539) - Async batch reprompt recovery loop (#528) - Batch target materialization — DB + disk parity (#527) - Halt workflow when upstream action fails all records (#523) - Unify batch and online guard evaluation, preflight, and lifecycle paths (#531)  **Observability** - Reprompt observability |
| `v0.1.16` | 2026-05-04 | High | ## What's Changed  ### Breaking Changes - **refactor: consolidate on_schema_mismatch into reprompt config** (#493) — Remove top-level `on_schema_mismatch` and `strict_schema` config keys. Schema conformance is now declared inside `reprompt: {on_schema_mismatch: reject/reprompt}`  ### Bug Fixes - **fix: return error dict on JSON parse failure instead of raising** (#492) — Return error dict instead of raising `VendorAPIError` on JSON parse failure in ollama cloud and openai online clients, enablin |
| `v0.1.14` | 2026-05-02 | High | ## v0.1.14  ### Breaking Changes - `read_target()` now validates `_state` field on all records (fail-closed). Pre-v0.1.14 target data requires deleting `agent_io/target/` and re-running. - `StorageBackend` subclasses must implement `_read_target_raw()` (template method)  ### Enhancements - **Record State Machine** — `RecordState` enum (8 states), `RecordEnvelope.transition()` as single lifecycle mutator, executor boundary reset, `derive_disposition()`, schema enforcement - **Unified processing** |
| `v0.1.13` | 2026-04-25 | High | ## Framework v0.1.13  ### Architectural Unification - Guard evaluation: 5 entry points → 1 unified evaluate() with GuardBehavior enum - RecordProcessor: 4 independent instantiations → single create() factory - Lineage building: deduplicated across exhausted/passthrough builders - Context builder: 235-line monolith → 4 focused functions (59-line orchestrator) - Dead code deleted: HistoricalNodeDataLoader, BatchProcessor, apply_observe_filter (-2992 lines)  ### Data Model Fixes - Fan-in merges all |
| `v0.1.12` | 2026-04-19 | High | ## v0.1.12 — 2026-04-19  ### Enhancement or New Feature * Self-reflection feedback strategy for reprompt retries * LLM critique escalation for stubborn reprompt failures  ### Under the Hood * Unified online and batch reprompt logic via shared helpers  ### Bug Fix * Drop directives now correctly exclude fields from passthrough wildcards * Pass compiled schemas to Groq, Cohere, and Mistral APIs * Retry transient OpenAI 400 JSON parsing errors * Cascade failures report as blocked instead of indepen |
| `v0.1.11` | 2026-04-13 | High | ## v0.1.11  ### Enhancement or New Feature  * Cross-workflow chaining: `--upstream` and `--downstream` CLI flags, virtual action injection, and orchestrator support for multi-workflow pipelines * FILE-mode record identity: tools receive full records with `node_id`, framework matches outputs to inputs by identity instead of heuristics (**breaking change** for FILE tools — use `record["content"]["field"]` instead of `record["field"]`) * Step-by-step progress display for sequential execution mode * |

## Dependency audit

- **Score**: 43/100
- **Total deps**: 44
- **Resolved**: 20
- **Unresolved**: 24
- **License conflicts**: 0
- **Warnings**: 23
- **Scanned**: 2026-05-11

## Citation

- HTML: https://www.freshcrate.ai/projects/agent-actions
- Markdown: https://www.freshcrate.ai/projects/agent-actions.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/agent-actions/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._