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.
npx agentmods add instructions/mohammad19974/bounded-relay/agents-mdgit clone --depth 1 https://github.com/mohammad19974/bounded-relayWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/mohammad19974/bounded-relay/agents-md)<a href="https://agentmods.dev/instructions/mohammad19974/bounded-relay/agents-md"><img src="https://agentmods.dev/badge/instructions/mohammad19974/bounded-relay/agents-md.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00715 | $0.00715 |
| Opus 5 | $0.00358 | $0.00358 |
| Sonnet 5 | $0.00143 | $0.00143 |
| Haiku 4.5 | $0.00072 | $0.00072 |
Grade A, and why
bounded-relay 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 3d 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.
How it starts
The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Repository Operating Rules
These rules apply to human and AI contributors working in this repository.
Product boundary
BoundedRelay is a local stdio MCP server that delegates bounded jobs from Claude
Code to the supported codex exec --json interface. It also provides optional,
additive SDD primitives for deterministic task routing and content-addressed
dual review.
Version 0.1 keeps two established Codex execution modes:
analyze: read-only execution in the validated source Git repository;proposal: opt-in execution in a disposable clean local clone, returning a validated binary patch that is never applied by this project.
The SDD router separately names two assignment lanes: codex and claude-host.
The latter is the current Claude Code session using its host-selected model. The
worker must never launch Claude or choose a Claude model.
Do not add direct source-worktree writes, automatic patch application, remote mutation, credential storage, model-quality claims, persistent jobs, or an audit ledger without an explicit design change and matching documentation.
Before changing code
- Read
README.md,docs/architecture.md, anddocs/security-model.md. - Inspect the current source, tests, package scripts, and working tree.
- Keep public documentation aligned with the actual implementation.
- Preserve the one-way topology: Claude Code → MCP worker → Codex CLI.
Implementation invariants
- Never interpolate task text into a shell command.
- Spawn executables with argument arrays and
shell: false. - Keep MCP stdout protocol-only; diagnostics belong on stderr.
- Keep analysis read-only by default.
- Proposal mode remains disabled unless
CCW_ENABLE_PROPOSALS=true. - Proposal work occurs only in the isolated clone and requires a clean, revision-pinned source.
- Validate paths, Git refs, patch size, changed-file count, and symlink boundaries before returning a proposal.
- Never apply, commit, push, publish, deploy, or modify a remote system.
- Never log or commit authentication material, prompts, job output, or local runtime state.
- Unknown Codex JSONL events must not crash a valid run merely because Codex added a new event type.
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.
- 3d ago First seen · 78 lines · 715 tokens per session scan A e30bef4875ee
bounded-relay AGENTS.md is an instructions file published in the GitHub repository mohammad19974/bounded-relay (1 stars, last pushed 4d ago), licensed MIT. It adds 715 tokens to every session, about $0.0036 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.
Other instructions, from other repositories
codex-trace AGENTS.md
Instructions for PixelPaw-Labs/codex-trace, covering claude.md, commands, dev, lint and format.
squarebox CLAUDE.md
Instructions for SquareWaveSystems/squarebox, covering claude.md, agent skills, issue tracker, triage labels and domain docs.
diaphora-mcp CLAUDE.md
Instructions for xTeardx/diaphora-mcp, covering diaphora mcp — claude code compatibility notes, project role, source map, non-negotiable data rules and large database defaults.
Osverse AGENTS.md
AGENTS.md instructions for Oswald-Hao/Osverse, covering osverse repository rules for coding agents and protected-branch workflow.
odin-codex-plugin AGENTS.md
AGENTS.md instructions for OutlineDriven/odin-codex-plugin, covering codex coding agent adherences, token-efficient cli output, tool hierarchy, core system & file ops and search & discovery.
cogitator AGENTS.md
Instructions for guilhermehto/cogitator, covering agents.md, layout, commands, demo / readme screenshot and conventions.