Skill Claude CodeCodex
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback.
Skill Claude CodeCodex
Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions.
Skill Claude CodeCodex
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always.
Skill Claude CodeCodex
A comprehensive verification system for Claude Code sessions. Use when verifying a Claude Code session's work before claiming it is complete.
Skill Claude CodeCodex
Use when planning a multi-step change before editing code.
Skill Claude CodeCodex
Use when creating new skills, editing existing skills, or verifying skills work before deployment.
Instructions file CodexOpenCode
AGENTS.md instructions for Ralph-Workflow/Ralph-Workflow, covering agents.md, scope, source of truth, priorities and non-negotiables.
Instructions file
Claude Code instructions for Ralph-Workflow/Ralph-Workflow: Claude-compatible agents working in this repository MUST follow the instructions in AGENTS.md. AGENTS.md is the single source of truth for project quality policy and project-specific agent behaviour. Do not duplicate or contradict its directives here.
Agent
The consolidated agent-registration model is built around five core abstractions: AgentCatalog owns all registrations and is the single injectable registry; AgentSupport is the unit of registration — one dataclass bundling a parser factory, a strategy factory, an AgentSpec, and an AgentConfig; AgentSpec captures the…
Agent
Every contributor and reviewer of any public-facing markdown file in this repository. If you edit, add, or approve a change to README.md, ralph-workflow/README.md, USERS.md (the canonical community directory after the 2026-07-07 docs cleanup), docs/, or the Sphinx operator manual, this page is your contract. The…
Agent
Single source of truth for all test strategy, rules, and patterns. Read before writing or modifying any test.
Agent
This document defines the only acceptable uses of # type: ignore[...] in the maintained Python package.
Agent
Agent "verification" from Ralph-Workflow/Ralph-Workflow, covering required verification (before pr/completion), canonical command, no unrelated-failure exemption — you find it, you fix it, fabrication guardrails and total test budget — 60 seconds, absolute and immutable.
Agent
ALL filesystem operations MUST go through the Workspace protocol. Direct open(), Path.readtext(), Path.writetext(), or any other pathlib/os filesystem call is FORBIDDEN in production pipeline and phase code. This is not a style preference — it is a hard constraint that enables testing.
Agent
See also: Agent Subsystem README for the discoverable entry point.
Agent
See Agent Subsystem README for the unified entry point.
Agent
All artifact submission side effects in Ralph Workflow must route through a single canonical backend entry point. This document explains the contract, what it protects, and how to stay on the right side of the audit.
Agent
Ralph treats an agent as broken when circumstantial evidence shows that it cannot produce meaningful LLM output. The recovery action is fallover to the next configured agent, rather than retrying the same agent.
Agent
This is the maintained closure map for the filesystem-proportional run plan. It records the public boundary, current evidence, and remaining gap for each criterion in .agent/PRODUCTCRITERIA.md. A GAP is not a waived requirement: it is the next implementation target. Re-prove the current baseline with.
Agent
Ralph's memory and process-lifecycle invariants, the bounded-subprocess audit that enforces them, and the known limitations operators must understand. The audit at ralph/testing/auditmcptimeout.py and its expansion (this document) make unbounded blocking calls structurally hard to reintroduce in the paths that matter.
Agent
This is the focused record for the speed-of-verified-delivery increment: commits a92a31941 through cf800b26f, plus the planning-template order correction. The stated (S) and recorded (R) criteria are complete and verified. The baseline (B) is frozen. The observed (O) criteria are staged but unrun: they require the…
Agent
See also: Agent Subsystem README for the unified entry point.
Agent
This document is the traceability map between each acceptance criterion (R1–R8) of the Trustworthy Idle Watchdog spec and the implementing module + pin test that enforces it. It is intentionally tutorial-free: no code, no implementation walkthrough — only the per-AC linkage. Drift between this document and the…