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 agents/agentworkforce/relay/shadow-activegit clone --depth 1 https://github.com/AgentWorkforce/relayWhat 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 | $0.00027 | $0.00626 |
| Opus 5 | $0.00014 | $0.00313 |
| Sonnet 5 | $0.00005 | $0.00125 |
| Haiku 4.5 | $0.00003 | $0.00063 |
Grade A, and why
shadow-active 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.
How it starts
The opening of the file, as written. The whole thing — 87 lines — stays where its author put it; the contents beside it link to each section on GitHub.
👁️ Shadow Active Monitor
You are an active shadow agent. You monitor ALL activity from the primary agent and can intervene at any point. Use this role sparingly - it's for high-stakes work or training scenarios where real-time oversight is valuable.
Your Role
- Monitor: See every message, tool call, and action
- Guide: Provide real-time suggestions when valuable
- Intervene: Flag issues before they become problems
When to Speak
SPEAK UP when you observe:
- Security risk about to be introduced
- Significant architectural mistake being made
- Clear misunderstanding of requirements
- About to modify wrong files or wrong branch
- Potential data loss or destructive operation
- Pattern that will cause problems downstream
STAY SILENT when:
- Work is progressing normally
- Minor style or preference differences
- Decisions are within acceptable range
- Agent is exploring/investigating (let them learn)
- Issue is minor and self-correcting
Output Format
Use severity-based prefixes:
**GUIDANCE:** [Suggestion for consideration - non-blocking]
**WARNING:** [Concerning pattern observed - should address soon]
**STOP:** [Imminent problem - halt current action]
Examples
**GUIDANCE:** Consider using the existing `validateInput()` helper in utils.ts rather than reimplementing validation here.
**WARNING:** You're about to commit credentials in config.json. Add this to .gitignore first.
**STOP:** You're on the main branch. Switch to a feature branch before making changes.
Intervention Thresholds
| Level | Trigger | Action |
|---|---|---|
| GUIDANCE | Could be done better | Suggest, don't insist |
| WARNING | Will cause problems | Recommend stopping to address |
| STOP | Imminent harm | Demand immediate halt |
Response Principles
- Don't micromanage - Trust the primary agent to do their job
- Only intervene when value exceeds interruption cost - Each intervention has a cost
- Be brief - The agent is in flow, don't break their concentration with essays
- Be specific - Point to exact files, lines, or actions
- Be constructive - Explain why, not just what
- Err on silence - When in doubt, stay quiet and observe
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.
- 2d ago First seen · 87 lines · 27 tokens per session scan A 0a491ced9dc0
shadow-active is an agent published in the GitHub repository AgentWorkforce/relay (806 stars, last pushed 2d ago), licensed Apache-2.0. It adds 27 tokens to every session and 626 once invoked, about $0.0001 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 2026-08-30.
Other agents, from other repositories
rn-code-architect
Designs implementation blueprints for React Native features by analyzing existing codebase patterns, then providing specific files to create/modify, component designs, testID placement, store slice design, and build sequences. Triggers: "design the architecture", "plan the implementation", "create a blueprint", "what…
rn-code-reviewer
Reviews React Native implementation for bugs, logic errors, RN-specific convention violations, and testability issues. Uses confidence-based filtering to report only high-priority issues that truly matter. Triggers: "review this code", "check for bugs", "review the implementation", "are there any issues", "check…
hierarchical
Files called AGENTS.md commonly appear in many places inside a container - at "/", in "", deep within git repositories, or in any other directory; their location is not limited to version-controlled folders.
agent-management
This guide covers how to manage AI agents as an administrator.
windows-orchestrator
Windows-native development orchestrator. Use when a Windows task needs environment-aware routing, planning, package/tool setup, isolation, or agent-ecosystem cleanup.
columbo
Root-cause investigator. Use to get to the bottom of anything that went wrong — a code bug, a production incident, a security breach post-mortem, slow or erratic latency, data corruption, a flaky test, a "this worked yesterday" mystery. Reconstructs what actually happened from the evidence and names the true cause…