spinje/pflow
16 subagents
Subagent
code-implementerImplement small, focused tasks in the pflow project: new functions/files, bug fixes, refactoring, or component integration. Caller MUST provide comprehensive context (requirements, file paths,…pflow-codebase-searcherSearch and navigate the pflow codebase. Use for: finding implementations, tracing data flows through CLI/runtime/nodes, understanding node lifecycle patterns, locating test coverage, resolving…review-agent-uxEvaluate every user-facing output (errors, warnings, success results, reports, CLI output) for AI agent actionability. pflow is agent-first — every message must help an AI agent diagnose and fix the…review-concurrency-safetyFind thread safety issues, resource lifecycle bugs, and Python-specific concurrency gotchas. Catches: shared mutable state across threads, ThreadPoolExecutor traps, deep copy crashes, copy semantics…review-feature-interactionsEnumerate how changes interact with existing features (batch, nested workflows, branching, caching, error handling, MCP, output/display).review-impact-completenessWhen a shared pattern is modified, find ALL consumers — including ad-hoc reimplementations that don't use shared code. The pattern behind the most subtle post-merge bugs.review-planReview implementation plans for structural integrity before coding begins. Catches: unverified assumptions, missing phases, wrong approach, ambiguous instructions, incomplete code path coverage,…REVIEW-PROTOCOLCLI tool for AI agents to build and discover automations. Multi-step tasks → reusable commands using LLM, shell, Code, HTTP, MCP, Agent nodes.review-silent-failuresFind operations that silently succeed when they should fail, warn, or produce empty results. The #1 post-merge bug category (30% of all fixes).review-simplicityJudge whether the FINAL integrated code is as simple as it should be — the dimension a correctness reviewer misses. Catches: emergent duplication across separately-implemented segments, interfaces…review-test-fidelityCheck that tests actually test the right thing — correct assertions, production-matching fixtures, behavior not implementation.review-validation-consistencyDetect drift between validation and runtime behavior. When runtime changes, validation must match — and vice versa. Catches: validators rejecting valid workflows, validators accepting invalid…task-orchestratorRuns ONE pflow task end to end in its worktree — plans (or implements a task-planner's plan), delegates phases, gates quality, writes task-review.md, opens the PR. Launched only by the main…task-phase-implementerImplements exactly the assigned phase(s) of a task's implementation-plan.md in the task's worktree — tests as it goes, logs substance, stops on ambiguity. Model per launch: sonnet mechanical, opus…task-plannerInvestigates ONE pflow task in its worktree, writes + self-reviews its implementation-plan.md, commits it on the feature branch, then stops. Launched only by the main orchestrator, on Opus.test-writer-fixerWrite new tests or fix failing tests in the pflow project. Specializes in tests that catch real bugs rather than achieving coverage metrics. Caller should provide: specific files/functions to test,…