Thrd Skill

Provision a dedicated inbox for your AI agent and manage email safely via thrd.email. Includes instant onboarding, inbound polling, reply/send (idempotent +...

Security Scan

Files, env vars, and runtime instructions are coherent with an agent-focused email integration (thrd.email); the skill only asks for the expected THRD_API_KEY and uses Python scripts to call the thrd API.

Purpose & Capability

Name/description (provision/manage an agent inbox via thrd.email) match the actual artifacts: Python scripts call https://api.thrd.email endpoints and requirements list python3 and THRD_API_KEY. The requested binaries and env var are appropriate for this purpose.

Instruction Scope

SKILL.md and scripts instruct the agent to run onboarding, checkout, openapi sync, and a poll daemon. These scripts only interact with the thrd.email API and write two local artifacts: a cache for openapi.json (.cache/openapi.json) and a cursor file (default .thrd_cursor). Onboarding intentionally redacts the API key unless --reveal-api-key is used. The instructions explicitly warn not to persist API keys. Note: onboarding can run without THRD_API_KEY (it provisions and returns a key), while other tools require it — SKILL.md documents this distinction.

Install Mechanism

There is no registry install spec, but SKILL.md suggests running `pip install -r requirements.txt`. requirements.txt lists only 'requests'. This is a low-risk, standard dependency; the install action is local and transparent. (Minor inconsistency: registry metadata listed no install, but the skill provides an install command in SKILL.md.)

Credentials

Only THRD_API_KEY is required, which is proportional to an API-based email tool. Two minor metadata inconsistencies: the registry 'primary credential' field is empty despite THRD_API_KEY being the primary secret, and SKILL.md correctly notes onboarding does not require the env var while other scripts do. No other unrelated secrets or config paths are requested.

Persistence & Privilege

The skill does not request elevated privileges or permanent platform presence (always:false). It writes only its own cache and cursor files and does not attempt to modify other skills or global configs. Autonomous invocation (model invocation allowed) is the platform default and not by itself concerning here.

Assessment

This skill appears to do what it says: provision and operate a dedicated agent inbox on thrd.email. Before installing, consider the following: (1) Keep THRD_API_KEY in a secure runtime secret store (the scripts advise this); only use --reveal-api-key in a trusted terminal. (2) The poll daemon and openapi sync write small local files (.thrd_cursor and .cache/openapi.json) — ensure the agent runtime can accept these files and their locations are acceptable. (3) SKILL.md includes a pip install step (requests only); confirm you want to run pip in the agent environment. (4) Metadata minor inconsistencies exist (no registry install spec, primary credential not declared), but they are not significant for security. If you need stronger assurance, ask the publisher for a provenance/homepage link or run the scripts in an isolated environment and inspect network traffic to ensure they only contact api.thrd.email.

Thrd Email Skill

This skill helps you create and operate an isolated inbox for an AI agent using thrd.email, without connecting your personal inbox.

Safety by default: don't connect your primary inbox to an agent; use a dedicated agent inbox.

Workflows

Sync API Contract (Recommended Before Tool Use)

To avoid stale assumptions, refresh the OpenAPI contract and read info.version:

python3 scripts/openapi_sync.py
python3 scripts/openapi_sync.py --print-version

This uses HTTP cache validators (ETag/Last-Modified) and only re-downloads when changed.

Provision a New Email Account

To create a new email account, run the onboarding script:

python3 scripts/onboard.py --agent-name "My Agent" [--tenant-name "My Company"]

This prints a JSON payload to stdout with the new inbox data and a redacted API key by default.

If you explicitly need the one-time raw key in a trusted terminal:

python3 scripts/onboard.py --agent-name "My Agent" --reveal-api-key

Security note: Do not write your API key to disk. Store it in your runtime's secret manager and set THRD_API_KEY as an environment variable. (The rest of the tools require THRD_API_KEY; onboarding does not.)

Upgrade Plan (Billing)

To start paid billing for your current tenant, use the checkout script:

python3 scripts/checkout.py <plan_name>

Plans:

• sandbox -> Sandbox Starter (9 EUR/month, raises Tier 1 monthly limit from 100 to 2,000 emails)
• limited -> Tier 2
• verified -> Tier 3

Forward the resulting Stripe URL to your human owner for payment.

Human Claiming (Verification)

Tier 3 (Verified Outbound) requires a responsible human linked via X.

• Start the flow: POST /v1/claim/x/start
• Forward the claim_url to your human owner.
• Check status: GET /v1/claim/x/status

Proof of Reasoning (PoR)

Cold outbound (Tier 3) may require a reasoning challenge to prevent spam.

• If you receive a 428 por_required error, solve the logical challenge provided in the response.
• Re-send the request with por_token and por_answer.

Manage Emails and Track Delivery

For detailed API usage (polling, sending, replying, trust scores, and checking delivery status), see references/api.md. Note: replies use reply-all behavior by default: they preserve historical CC and keep recipients from the latest inbound To line so participants are not dropped. Tier2+ may add CC via cc[]. In Tier1, cc[] may only contain addresses already present in that thread's CC history. Security note: when Prompt Shield marks an inbound email as high-risk, Tier2/3 flows may require creating a short-lived security_ack_token (POST /v1/security/ack) before reply/send. Quota note: use GET /v1/usage to monitor monthly usage (used, remaining, state, reset_at) and avoid hitting hard limits mid-run.

Wake-Up Strategy (Recommended)

Many LLM runtimes do not reliably maintain background polling. Use wake webhooks when possible:

• Configure webhook: PUT /v1/wake/webhook
• Read status: GET /v1/wake/webhook
• Disable webhook: DELETE /v1/wake/webhook

THRD sends signed inbox.pending pings, then your runtime should immediately pull with GET /v1/events and ACK.

Fallback when webhooks are not available:

python3 scripts/poll_daemon.py --cursor-file .thrd_cursor

This keeps pull-based delivery alive without requiring a public webhook endpoint.

Tools

• scripts/onboard.py: Instant provisioning of a new email inbox.
• scripts/checkout.py: Generate a Stripe Checkout URL for upgrades.
• scripts/openapi_sync.py: Refresh/cache latest OpenAPI and read current info.version.
• scripts/poll_daemon.py: Fallback long-poll daemon for runtimes without wake webhook support.