lobster/CHANGELOG.md

Changelog

All notable changes to Lobster will be documented in this file.

Unreleased

• Add first-class openclaw.agent workflow turns with configured agent, session, model, thinking, and timeout selection delegated to OpenClaw. Thanks to @Stoff81 (Issue #117).

2026.6.11

• Add command-level ctx.requestInput(...) for CLI/tool/SDK pipeline commands, with state-backed same-command resume, bounded command-input replay, and workflow pipeline: propagation (Issue #101).
• Warn when LLM usage records an unknown or missing model ID or invalid LOBSTER_LLM_PRICING_JSON, keeping zero-cost fallback behavior visible for cost_limit users. Thanks to @KrasimirKralev (Issue #107).
• Require Node.js 22 or newer for the npm package, matching release CI.
• Write Lobster state files atomically while preserving restricted file modes, preventing truncated resume/session state after process termination. Thanks to @KrasimirKralev (Issues #108, #109, PR #110).
• Harden LLM cache files, diff snapshots, and approval ID indexes against truncated JSON after process termination. Disposable cache/snapshot corruption now recovers as a miss, authoritative resume state still surfaces malformed JSON, and approval short-ID indexes are published atomically without overwriting existing mappings. Thanks to @TurboTheTurtle (Issues #111, #112, #113, PR #114).
• Fix timeout_ms + retry so per-attempt timeouts retry as documented while external workflow cancellation still stops immediately. Thanks to @KrasimirKralev (PR #106).

2026.5.22

• Memoize Ajv schema compilation for repeated validation paths to avoid retained SchemaEnv/closure growth in long-running processes. Thanks to @KrasimirKralev (PR #98) and @cmi525 (Issue #96).
• Improve workflow resume compatibility for stateKey naming by accepting both workflow_resume_ and workflow-resume_ prefixes, including cleanup against the resolved on-disk key. Thanks to @brownetw-ai (PR #4).
• Add per-step workflow retry policies (max, backoff, delay_ms, max_delay_ms, jitter) with retry-aware stderr logs and dry-run visibility. Thanks to @scottgl9 (PR #84).
• Add optional approval identity constraints for workflow gates (approval.initiated_by, approval.required_approver, approval.require_different_approver) with resume-time enforcement via LOBSTER_APPROVAL_APPROVED_BY and envelope metadata for integrations. Thanks to @coolmanns (Issue #44).
• Clarify pipeline: vs run: usage for llm.invoke / llm_task.invoke in workflow files, and add regression coverage to ensure stdin: $step.stdout is forwarded as LLM artifacts for llm_task.invoke pipeline steps. Thanks to @RatkoJ (Issue #41).
• Add lobster graph workflow visualization with mermaid (default), dot, and ascii outputs, including step-type nodes, stdin data-flow edges, conditional dependency labels (when/condition), approval-gate diamond shapes, and --args-json label resolution support. Thanks to @vignesh07 (Issue #53).
• Add workflow composition via workflow: + workflow_args, including recursive sub-workflow execution, cycle detection, and dry-run visibility for workflow steps. Sub-workflow approval/input halts are rejected with resume-state cleanup. Thanks to @scottgl9 (PR #73).
• Add per-step on_error workflow policies (stop|continue|skip_rest) for partial-failure recovery, with structured step error fields (error, errorMessage) for condition-based branching. Thanks to @scottgl9 (PR #72).
• Add per-step workflow timeout_ms handling, including timeout-triggered aborts, SIGKILL for timed shell steps, and dry-run annotations. Thanks to @scottgl9 (PR #74).
• Add workflow condition comparison operators <, <=, >, and >= with strict numeric semantics (booleans/null do not coerce), including mixed boolean-expression support with &&/||. Thanks to @scottgl9 (PR #71).
• Add workflow-level LLM cost tracking with _meta.cost summaries, per-step usage attribution, and optional cost_limit controls with warn/stop actions (plus custom pricing via LOBSTER_LLM_PRICING_JSON). Thanks to @scottgl9 (PR #70).
• Add parallel workflow steps with branch fan-out, wait: all|any, block-level timeout support, and branch result references in downstream steps. Thanks to @scottgl9 (PR #69).
• Add for_each workflow steps for per-item sub-step execution over arrays, including loop-scoped vars (item_var/index_var), optional batch_size + pause_ms, and collected iteration outputs for downstream steps. Thanks to @scottgl9 (PR #68).
• Add pipe-based template filters (for example upper, length, join, default, date) for the template command with quote-aware filter parsing and chain evaluation. Thanks to @scottgl9 (PR #67).

2026.4.6

• Add workflow file support for .lobster, YAML, and JSON, including workflow args/env, native pipeline steps, and shell-safe LOBSTER_ARG_* inputs.
• Add structured input pauses with ask, workflow input, needs_input, and lobster resume --response-json '{...}' for resumable human-in-the-loop flows.
• Add richer workflow condition expressions with !, ==, !=, &&, ||, and parentheses.
• Export the embeddable runtime via @clawdbot/lobster/core so Lobster can run in-process inside OpenClaw and other hosts.
• Add generic llm.invoke adapters, openclaw.invoke --each, and keep clawd.invoke as a supported alias.
• Add compact state-backed workflow/pipeline resume tokens, safer resume validation, and hardened approval ID handling.
• Improve dry-run and shell interoperability with exec --stdin raw|json|jsonl, approve --preview-from-stdin --limit N, and better template/shell-variable preservation.
• Improve Windows CLI/build compatibility and fix quoted-argument parser edge cases.

2026.1.21-1

• Published release (pre-changelog).

2026.1.21

• Initial published release (pre-changelog).