hydraflow AGENTS.md

Repository instructions describing HydraFlow's agent roles, prompt contracts, supported command-line runtimes, and required output markers. A runtime is the command-line agent program that reads instructions and returns text.

In plain words
What is it for?
Use them when editing HydraFlow agent prompts or runtime integration. They guide planning, implementation, review, human-approval stages, runtime selection, and stable output formatting.
Why use it?
They help keep different agents interchangeable and ensure HydraFlow can parse each stage's output reliably. They also document constraints that are not obvious from the prompt text alone.

Instructions file for CodexOpenCode

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/t-rav/hydraflow/agents-md
Clone the repo
git clone --depth 1 https://github.com/T-rav/hydraflow

Made for: Codex, OpenCode.

Per session 5,037 This file is loaded in full into every session.
When invoked 5,037 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.05037 $0.05037
Opus 5 $0.02518 $0.02518
Sonnet 5 $0.01007 $0.01007
Haiku 4.5 $0.00504 $0.00504

Measured 2d ago against content hash 8a771462115b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hydraflow AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

AGENTS.md · 474 lines

How it starts

The opening of the file, as written. The whole thing — 474 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md

Canonical reference for every agent role in HydraFlow. Each section documents:

  • Role — what the agent does and its constraints
  • Prompt structure — required sections and ordering
  • Output contract — exact markers the system parses
  • Key rules — non-negotiable constraints

Source of truth for prompt intent. Prompt text lives in the runner files; this document explains why each prompt is structured the way it is.


Agent Runtimes

HydraFlow is runtime-agnostic. The same prompt contracts work with any agent CLI that can read a prompt from stdin / a flag and write text to stdout. The output markers (PLAN_START, VERDICT:, SUMMARY:, etc.) are plain text and impose no tool-specific requirements.

Supported runtimes

Runtime CLI invocation Config key
Claude Code claude -p "<prompt>" planner_tool = "claude"
OpenAI Codex codex "<prompt>" planner_tool = "codex"
Pi.dev pi "<prompt>" planner_tool = "pi"

Each stage (plan, implement, review, HITL) can use a different runtime:

HYDRAFLOW_PLANNER_TOOL=claude
HYDRAFLOW_IMPLEMENT_TOOL=codex
HYDRAFLOW_REVIEW_TOOL=claude
HYDRAFLOW_HITL_TOOL=pi

What the runtime must support

For any runtime to work with HydraFlow:

  1. Stdin / flag prompt ingestion — the prompt is passed as a positional argument or --prompt flag.
  2. Filesystem access — the agent must be able to read files in the working directory (read-only for the planner, read-write for others).
  3. Shell tool accessmake lint, make quality, git commands must be executable from within the agent session (for implementer and reviewer).
  4. Plain-text stdout — output markers are parsed from raw stdout/stderr; no structured JSON response format is required.

Output marker stability

The following markers are parsed by HydraFlow's Python code and must not be renamed without a coordinated update to the corresponding runner:

Read the full file on GitHub · 474 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 474 lines · 5,037 tokens per session scan A 8a771462115b

Subscribe to this mod's changes

hydraflow AGENTS.md is an instructions file published in the GitHub repository T-rav/hydraflow (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 5,037 tokens to every session, about $0.0252 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.