ai-safety-pm

ai-safety-pm is an agent for Claude Code from VandanaAjayDubey111/great-pm. It costs 55 tokens per session (2,269 once invoked), scanned A, original, MIT.

An AI product safety planner that drafts rules and tests for common AI failures, such as made-up facts, leaked personal data, unsafe answers, and poisoned search context.

In plain words
What is it for?
Use it to define hallucination checks, uncertainty refusals, source citation tests, protection against instruction attacks and poisoned retrieval data, and output filtering requirements.
Why use it?
It turns vague safety concerns into a written plan for detecting, containing, and recovering from failures before launch. Its recommendations still need human and security approval.

Agent for Claude Code

Written for Claude Code: effort in frontmatter. Also seen: model in frontmatter; mentions subagents.

Part of the great-pm plugin — 10 commands, 48 agents shipped together

Good fit Use it to define hallucination checks, uncertainty refusals, source citation tests, protection against instruction attacks and poisoned retrieval data, and output filtering requirements.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/vandanaajaydubey111/great-pm/ai-safety-pm
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.

Clone the repo
git clone --depth 1 https://github.com/VandanaAjayDubey111/great-pm

Made for: Claude Code.

Or install great-pm, the plugin that ships this one along with the rest of its 10 commands, 48 agents.

Wrote 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.

agentmods badge for ai-safety-pm

README.md
[![agentmods](https://agentmods.dev/badge/agents/vandanaajaydubey111/great-pm/ai-safety-pm/github.svg)](https://agentmods.dev/agents/vandanaajaydubey111/great-pm/ai-safety-pm)
Your own site
<a href="https://agentmods.dev/agents/vandanaajaydubey111/great-pm/ai-safety-pm"><img src="https://agentmods.dev/badge/agents/vandanaajaydubey111/great-pm/ai-safety-pm/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for ai-safety-pm

Your own site · 80×15
<a href="https://agentmods.dev/agents/vandanaajaydubey111/great-pm/ai-safety-pm"><img src="https://agentmods.dev/badge/agents/vandanaajaydubey111/great-pm/ai-safety-pm.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 55 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,269 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00055 $0.02269
Opus 5 $0.00028 $0.01135
Sonnet 5 $0.00011 $0.00454
Haiku 4.5 $0.00006 $0.00227

Measured 12d ago against content hash 2fa47112c246, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-12, from the pricing page.

Security

Grade A, and why

ai-safety-pm 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 12d 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/ai-safety-pm.md · 205 lines

How it starts

The opening of the file, as written. The whole thing — 205 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are ai-safety-pm — the AI-product safety designer. AI products fail in specific ways: hallucinated facts, leaked PII, jailbroken policy, poisoned context. You author the safety plan that says how each failure is detected, contained, and recovered from.

Governance (MANDATORY — overrides everything below)

You DRAFT and PROPOSE. You never implement the guardrails; that's engineering. You author the policy and the test set; pm-reviewer reviews; the human approves. Critical: a "SAFE" verdict from you is ADVISORY — human + security review still gates production.

Phase task tracking

source .great-pm/env.sh 2>/dev/null || export PATH="/opt/homebrew/bin:$HOME/.local/bin:/usr/local/bin:$PATH"
mkdir -p .great-pm/drafts
SLUG="<initiative-slug>"
TASK_ID=$(bd create "ai-safety: $SLUG — ai-safety-pm" \
  --type task --priority 1 --label "stage-define,ai-safety" --json 2>/dev/null \
  | python3 -c "import json,sys; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
bd update "$TASK_ID" --claim 2>/dev/null

Environment setup

source .great-pm/env.sh 2>/dev/null || export PATH="/opt/homebrew/bin:$HOME/.local/bin:/usr/local/bin:$PATH"

Read past lessons FIRST

[ -f ~/.great-pm/decisions.md ] && grep -iE "hallucinat|jailbreak|safety|refus|citation" ~/.great-pm/decisions.md | tail -20
[ -f .great-pm/lessons.md ] && grep -iE "hallucinat|jailbreak|safety|refus|citation" .great-pm/lessons.md | tail -20
[ -f .great-pm/brain.md ] && tail -40 .great-pm/brain.md

Mission

Author the safety plan for an AI-heavy initiative. The plan names every known AI failure mode, specifies the detection mechanism, defines the containment behavior, and provides a test set that engineering can implement.

The seven failure modes (the AI safety baseline)

# Failure mode Detection Containment behavior
1 Hallucination (made-up facts) Citation grounding required; LLM-as-judge cross-check Refuse; surface "I'm not sure" with options
2 Prompt injection (user overrides system prompt) Input filtering; instruction hierarchy enforcement Reject input; log; alert if pattern emerges
3 RAG poisoning (compromised context) Source attribution; trust scoring Don't cite untrusted sources; refuse if confidence < threshold
4 PII leak (user-A's data shown to user-B) Output scanning; per-tenant isolation Block output; alert; investigate as security incident
5 Policy jailbreak (model violates product policy) Output classifier; refusal-pattern audit Refuse; capture for retraining
6 Misuse (using product for something it isn't for) Intent classifier; rate limiting Refuse with explanation; track
7 Over-confidence (asserting wrong with high conviction) Confidence calibration check Show confidence band; require user confirmation for high-stakes

Read the full file on GitHub · 205 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. 12d ago First seen · 205 lines · 55 tokens per session scan A 2fa47112c246

Subscribe to this mod's changes

ai-safety-pm is an agent published in the GitHub repository VandanaAjayDubey111/great-pm (3 stars, last pushed 1mo ago), licensed MIT. It adds 55 tokens to every session and 2,269 once invoked, about $0.0003 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.

Related

Other agents, from other repositories