Reah Skill: Agent Card

Retrieve masked card info from Reah using an access key. Handles session generation, secure fetch, and decryption for agents automatically.

Security Scan

The skill's behavior roughly matches its purpose (fetching and decrypting card fragments), but there are metadata inconsistencies and origin/installation ambiguities you should resolve before trusting it with sensitive keys.

Purpose & Capability

The skill claims to retrieve masked card info from Reah and the included Node example implements a GraphQL call to https://agents.reah.com/graphql and local decryption — this is coherent with the description. However the package/registry metadata provided to the evaluator omits the REAH_AGENT_KEYS env var that the SKILL.md and README clearly require, creating an inconsistency between declared requirements and the runtime instructions.

Instruction Scope

SKILL.md limits network calls to the single Reah GraphQL endpoint, requires explicit user confirmation before reading REAH_AGENT_KEYS, and mandates masking/no-export of raw PAN/CVC. The example Node script enforces endpoint immutability and does the decryption locally. That scope is appropriate for the stated goal. Caveat: the example decrypts values in memory but does not show or save them — enforcement of masking/never-exposing card data is purely procedural (instructions), not enforced across the skill surface.

Install Mechanism

This is an instruction-only skill with an included reference script; there is no install spec that downloads remote artifacts. README suggests an npx install from a GitHub repo, but no install spec in the registry package. No remote download URLs or installers were found in the provided files.

Credentials

The skill expects sensitive REAH_AGENT_KEYS to be available (and the SKILL.md metadata lists REAH_AGENT_KEYS). That is proportionate to the function, but the registry metadata earlier reported 'Required env vars: none' — this mismatch is concerning. Also the README instructs adding a JSON mapping to REAH_AGENT_KEYS in agent env. Ensure the skill will only read keys after explicit per-read confirmation as required by SKILL.md and that the agent/platform enforces that confirmation flow rather than silently reading environment variables.

Persistence & Privilege

The skill does not request always:true, does not modify other skills or system settings, and has normal invocation privileges. Nothing requests elevated or permanent system presence.

What to consider before installing

This skill appears to implement the described Reah card access flow and keeps network activity constrained to agents.reah.com, but there are a few things to verify before installing: - Confirm provenance: the registry lists the source as unknown and SKILL.md/README point to a GitHub install; verify the skill's origin (official Reah repo) before adding it to an agent that will handle card keys. - Metadata mismatch: SKILL.md requires REAH_AGENT_KEYS but the registry metadata you saw did not declare this — ask the publisher to correct the manifest so required env vars are explicit. - Review the confirmation flow: SKILL.md requires an explicit per-read confirmation for REAH_AGENT_KEYS. Ensure your agent platform actually prompts and prevents silent env reads. - Least privilege: store REAH_AGENT_KEYS only where necessary, rotate keys regularly as advised, and prefer short-lived keys if Reah supports them. - Code audit: the included Node example decrypts sensitive material in memory (but doesn't print it). If you plan to enable autonomous use, audit how the agent will use decrypted values and ensure it will only return masked/redacted card parts as specified. If you cannot verify the skill's source or guarantee the per-read confirmation behavior, treat this skill as higher risk and avoid installing it in environments with real card keys.

Reah Skill

This skill is organized by modules.

Modules

reah_card

Handle Reah card key flow for agents.reah.com.

This module handles two tasks.

Task 1: Ask for access key

If user did not provide an access key, ask user with this exact message:

To continue, I need your Reah card access key.

You can get it from agents.reah.com:
- Open your card
- Click "Generate agent key"

Paste it here and I'll securely fetch your card details.

If the workflow uses REAH_AGENT_KEYS from environment:

• MUST ask for manual confirmation before each key read, even within the same conversation.
• MUST NOT reuse prior confirmation.
• Use this exact confirmation message:

I can read the access key from REAH_AGENT_KEYS for this request.

Please confirm I should proceed with this key read now.

• After confirmation, remind the user to rotate access keys periodically.

Do not proceed to Task 2 before key is provided or key-read confirmation is granted.

Task 2: Get and decrypt card info

Example script (reference only)

Use the example script below as reference for the full process:

node {baseDir}/scripts/get-card-info-example.mjs \
  --access-key "<accessKey>"

This script includes all steps in one place:

• generate sessionId / secretKey
• request individualCardByAccessKey(accessKey, sessionId) from https://agents.reah.com/graphql
• decrypt encryptedPan and encryptedCvc

This script is for reference only. It intentionally ends after decryption and does not output raw pan/cvv.

Script Files

• {baseDir}/scripts/get-card-info-example.mjs

Security Constraints

• MUST use only the default Reah GraphQL endpoint: https://agents.reah.com/graphql.
• MUST NOT allow endpoint override.
• MUST NOT allow custom headers, cookies, or bearer authentication overrides.
• MUST NOT send card data to any external endpoint.
• MAY read access key from REAH_AGENT_KEYS only after explicit manual user confirmation for the current read.
• MUST require manual confirmation before every key read from REAH_AGENT_KEYS.
• MUST remind users to rotate access keys periodically whenever key-read confirmation is requested.
• MUST NOT expose full access key in any user-facing response.
• MUST NOT expose raw secretKey in any user-facing response.
• MUST NOT return raw card info in any user-facing response. Card info part A MUST be masked (for example **** **** **** 1234) and card info part B MUST be redacted.

Error Handling

• If access key is invalid, ask user to regenerate a new agent key and retry.
• If request fails or times out, retry once automatically with the same inputs.
• If retry still fails, ask user to check network/auth status and provide a fresh key.