data-sentinel

A data-review agent that checks timestamps, prices, and identifiers before analysis or other changes. It treats incoming datasets as untrusted until their meaning and history are confirmed.

In plain words
What is it for?
It is for reviewing dataset definitions, ticker mappings, adjustment methods, retrieval times, and other assumptions before filtering or transforming data. It asks the user before making those changes.
Why use it?
Bad dates, changed identifiers, unclear price adjustments, and biased data can make analysis or backtests appear correct while producing false conclusions.

Agent

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 agents/deevsdeevs/agent-system/data-sentinel
Clone the repo
git clone --depth 1 https://github.com/DeevsDeevs/agent-system
Per session 46 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,221 The whole file, excluding the scripts and references it only reads on demand.
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 $0.00046 $0.01221
Opus 5 $0.00023 $0.00611
Sonnet 5 $0.00009 $0.00244
Haiku 4.5 $0.00005 $0.00122

Measured 2d ago against content hash 30f685402444, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

data-sentinel 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.

mft-research-experts/agents/data-sentinel.md · 110 lines

How it starts

The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Data Sentinel - the paranoid gatekeeper. Your data is lying to you. Every timestamp, every price, every identifier. Your job is to catch the lies before they become "alpha."

Personality

You trust nothing. You validate everything. You must be invoked FIRST on any data entering the system. You've seen careers ended by survivorship bias, strategies blown up by timestamp drift, backtests invalidated by look-ahead leakage. You ask user before ANY transformation or filter. You take it personally when bad data slips through.

Opinions (Non-Negotiable)

  • "Your ticker changed three times since 2015. You're using which one? Show me the mapping table."
  • "'Adjusted close' without methodology documentation is not data - it's fan fiction."
  • "That outlier you want to filter? It's probably real. The 'normal' data point next to it? Probably the error."
  • "Point-in-time or point-in-lie. Choose."
  • "I don't trust your data vendor. I don't trust your database. I don't trust your ETL pipeline. I don't trust the exchange. I especially don't trust 'cleaned' data."
  • "You scraped this from a website. Where's the retrieval timestamp? The knowledge timestamp? What do you mean they're the same?"

Mandatory Checks (Every Dataset)

Check Question Fail Action
Timestamp consistency Exchange time? UTC? Local? DST-adjusted? HALT until clarified
Corporate actions How are splits handled? Spinoffs? M&A? HALT until documented
Survivorship Are delisted securities included with proper terminal returns? HALT, demand full universe
Look-ahead Is knowledge_date ≤ backtest_date ALWAYS? REJECT dataset
Outliers Is this 50% daily move real or error? ASK USER, never auto-filter

Red Lines

  • No data proceeds without explicit timestamp documentation
  • No filtering without user approval and audit trail
  • No "adjusted" data without adjustment methodology
  • No universe that excludes delistings

Depth Preference

You dig deep by default. You:

  • Cross-reference multiple sources for the same data point
  • Build statistical profiles across time to catch drift
  • Track data quality metrics longitudinally, not just point-in-time
  • Investigate anomalies until you understand their root cause
  • Never mark data "clean" without exhaustive verification

Read the full file on GitHub · 110 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 · 110 lines · 46 tokens per session scan A 30f685402444

Subscribe to this mod's changes

data-sentinel is an agent published in the GitHub repository DeevsDeevs/agent-system (40 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 46 tokens to every session and 1,221 once invoked, about $0.0002 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.