Squid

### Robust JSON Extraction - New `json-extract.ts` module: `parseAgentOutput()` extracts JSON from markdown fences (`` ```json ... ``` ``), embedded objects in prose, and raw JSON - All 3 adapters (OpenClaw, Claude Code, OpenCode) now use `parseAgentOutput()` instead of fragile `JSON.parse(stdout)` - Fixes the production issue where OpenClaw agents return JSON wrapped in markdown or prose instead of raw JSON - 16 unit tests covering all extraction patterns ### Claude Code Sub-Agent Support - `agentId` now works with the `claude-code` adapter — passed as `--agent <name>` to the CLI - Targets Claude Code sub-agents defined in `.claude/agents/<name>.md` - `agentId` is no longer OpenClaw-only — works across `claude-code` and `openclaw` adapters ### Parser Validation - Warning when `model:` is set with `runtime: subagent` — model is ignored for OpenClaw sub-agents (model is configured per agent in OpenClaw, not in the pipeline) ### OpenClaw Adapter Cleanup - Removed dead `OPENCLAW_TOKEN` / `OPENCLAW_GATEWAY_TOKEN` / `CLAWD_TOKEN` env var handling — the `openclaw` CLI never reads these; it authenticates via `openclaw config` stored credentials - Removed `token` config field from `OpenClawConfig` interface ### Documentation - Updated `docs/adapters.md`: fixed `agentId` support table (now "Yes" for Claude Code), corrected OpenClaw auth to CLI-only (no HTTP mode), updated comparison table - Updated SKILL.md with full adapter guide: when to use which adapter, `SQUID_AGENT` env var, `agentId` cross-adapter table, `model:` behavior per runtime - Added raw GitHub URLs for all doc files in SKILL.md Documentation Links section - Clarified `${ref}` interpolation in transforms (R7) vs bare `$ref` in `when:` conditions

squid-pipeline 0.1 — Initial release - Introduces Squid pipeline YAML workflow support, including multi-agent orchestration, gates, parallel execution, loops, branching, and structured approvals. - Enforces 10 core rules for safe, deterministic, and testable pipelines (e.g., mandatory gates before side effects, onError handling, spawn output specification, retry on network, loop limits). - Requires use of provided reference files (step-types, patterns, testing) for accurate pipeline and test generation. - Supports agent adapters (OpenClaw, Claude Code, OpenCode), sub-pipelines, and human-in-the-loop steps. - Provides install instructions, pipeline structure guide, and example patterns for rapid onboarding. - Mandates comprehensive test coverage for pipelines, including happy path, rejections, failures, and restart exhaustion.