✅ Concepts

Approval Workflows

Name: Agent Secret Store
Price: 49 USD

Human-in-the-loop gates for sensitive and critical credentials. Agents request access; humans approve or deny in real time.

Access tiers

Every secret is assigned an access tier that controls the approval requirement when an agent requests a scoped token to read it.

Standard

Auto-approved

Tokens are issued instantly. No human approval required. Suitable for non-sensitive operational credentials like internal API keys.

Examples

› Internal service URLs
› Development API keys
› Feature flags
› Non-secret configuration

Sensitive

1 approval required

One designated approver must approve within the request TTL. On approval, the token is issued automatically.

Examples

› Production API keys
› OAuth client secrets
› Database credentials
› LLM provider keys

Critical

1 approval required

The tenant owner must approve before the token is issued. Use for credentials with high blast radius or irreversible side effects.

Examples

› Payment processor keys
› Infrastructure credentials
› Master database passwords
› Signing keys

Approval flow

Agent requests scoped token— scopes + TTL

↓

Vault checks secret tier— standard → auto-issue | sensitive/critical → queue

↓

Approvers notified— Dashboard notification + email

↓

Approvers review + approve/deny— Dashboard UI

↓

Token issued to agent— Or denied — request logged in audit trail either way

Approval thresholds

Approval requirements are derived from the highest access tier matched by the requested scopes. Sensitive and critical secrets always require an explicit human approval before the token is issued.

Matched tier	Token behavior	Typical use
standard	Auto-issued	Low-risk operational credentials
sensitive	Requires 1 approval	Production model provider keys and OAuth client secrets
critical	Requires 1 approval	Payment, infrastructure, signing, and production database credentials

Configuring access tiers

Set the access tier when you store or update a secret. Future token requests that match sensitive or critical secrets will return an approval ID instead of a token until a human approves the request.

Shell

# Access tiers are assigned per secret.
ass secrets set production/gemini/GEMINI_API_KEY "$GEMINI_API_KEY" \
  --type api_key \
  --tier sensitive

ass secrets set production/stripe/STRIPE_SECRET_KEY "$STRIPE_SECRET_KEY" \
  --type api_key \
  --tier critical

Handling approvals in code

approval_flow.py

import asyncio
import os
from agentsecretstore import AgentVault, ApprovalRequiredError

SCOPE = "secrets:read:production/stripe/*"
SECRET_PATH = "production/stripe/STRIPE_SECRET_KEY"

async def wait_for_approval(vault: AgentVault, approval_id: str) -> None:
    while True:
        status = await vault.get_approval_status(approval_id)
        if status.status == "approved":
            return
        if status.status in {"denied", "expired"}:
            raise RuntimeError(f"Approval {status.status}: {approval_id}")
        await asyncio.sleep(5)

async def main():
    async with AgentVault(agent_key=os.environ["ASS_AGENT_KEY"]) as vault:
        try:
            await vault.request_token(scopes=[SCOPE], ttl="5m")
        except ApprovalRequiredError as exc:
            print(f"Approval required. Request ID: {exc.approval_id}")
            print("Approvers review the request in the dashboard.")
            await wait_for_approval(vault, exc.approval_id)
            await vault.request_token(scopes=[SCOPE], ttl="5m")

        secret = await vault.get_secret(SECRET_PATH)
        print(secret)

asyncio.run(main())

Non-blocking pattern

For production agents, don't block on approval. Instead, queue the task and resume after get_approval_status() returns an approved status.

Checking approval status

Approval decisions are made by humans in the dashboard. Agents receive an approval ID, then poll until the request is approved, denied, or expired.

Poll status

Shell

# Agents can poll an approval request by ID.
curl -X GET https://api.agentsecretstore.com/v1/approvals/agent/0b2e1c98-5ef3-4d2c-a7aa-45a91a18a201 \
  -H "Authorization: Bearer $ASS_AGENT_KEY"

Status payload

JSON

{
  "id": "0b2e1c98-5ef3-4d2c-a7aa-45a91a18a201",
  "tenant_id": "0e92b4ed-8e0e-4d83-b35a-a936fda75834",
  "agent_id": "0fd78c7f-53dd-4a87-8a0c-74b5a4c9a2db",
  "status": "pending",
  "requested_scopes": ["secrets:read:production/stripe/*"],
  "access_tier": "critical",
  "required_approvals": 1,
  "current_approvals": 0,
  "approvers": [],
  "reason": null,
  "denial_reason": null,
  "expires_at": "2026-06-01T11:00:00Z",
  "created_at": "2026-06-01T10:30:00Z",
  "resolved_at": null
}

Human action

Text

# Human approvers approve or deny requests in the dashboard:
https://agentsecretstore.com/dashboard/approvals

✅ Concepts

Approval Workflows

Human-in-the-loop gates for sensitive and critical credentials. Agents request access; humans approve or deny in real time.

Access tiers

Every secret is assigned an access tier that controls the approval requirement when an agent requests a scoped token to read it.

Standard

Auto-approved

Tokens are issued instantly. No human approval required. Suitable for non-sensitive operational credentials like internal API keys.

Examples

› Internal service URLs
› Development API keys
› Feature flags
› Non-secret configuration

Sensitive

1 approval required

One designated approver must approve within the request TTL. On approval, the token is issued automatically.

Examples

› Production API keys
› OAuth client secrets
› Database credentials
› LLM provider keys

Critical

1 approval required

The tenant owner must approve before the token is issued. Use for credentials with high blast radius or irreversible side effects.

Examples

› Payment processor keys
› Infrastructure credentials
› Master database passwords
› Signing keys

Approval flow

Agent requests scoped token— scopes + TTL

↓

Vault checks secret tier— standard → auto-issue | sensitive/critical → queue

↓

Approvers notified— Dashboard notification + email

↓

Approvers review + approve/deny— Dashboard UI

↓

Token issued to agent— Or denied — request logged in audit trail either way

Approval thresholds

Approval requirements are derived from the highest access tier matched by the requested scopes. Sensitive and critical secrets always require an explicit human approval before the token is issued.

Matched tier	Token behavior	Typical use
standard	Auto-issued	Low-risk operational credentials
sensitive	Requires 1 approval	Production model provider keys and OAuth client secrets
critical	Requires 1 approval	Payment, infrastructure, signing, and production database credentials

Configuring access tiers

Set the access tier when you store or update a secret. Future token requests that match sensitive or critical secrets will return an approval ID instead of a token until a human approves the request.

Shell

# Access tiers are assigned per secret.
ass secrets set production/gemini/GEMINI_API_KEY "$GEMINI_API_KEY" \
  --type api_key \
  --tier sensitive

ass secrets set production/stripe/STRIPE_SECRET_KEY "$STRIPE_SECRET_KEY" \
  --type api_key \
  --tier critical

Handling approvals in code

approval_flow.py

import asyncio
import os
from agentsecretstore import AgentVault, ApprovalRequiredError

SCOPE = "secrets:read:production/stripe/*"
SECRET_PATH = "production/stripe/STRIPE_SECRET_KEY"

async def wait_for_approval(vault: AgentVault, approval_id: str) -> None:
    while True:
        status = await vault.get_approval_status(approval_id)
        if status.status == "approved":
            return
        if status.status in {"denied", "expired"}:
            raise RuntimeError(f"Approval {status.status}: {approval_id}")
        await asyncio.sleep(5)

async def main():
    async with AgentVault(agent_key=os.environ["ASS_AGENT_KEY"]) as vault:
        try:
            await vault.request_token(scopes=[SCOPE], ttl="5m")
        except ApprovalRequiredError as exc:
            print(f"Approval required. Request ID: {exc.approval_id}")
            print("Approvers review the request in the dashboard.")
            await wait_for_approval(vault, exc.approval_id)
            await vault.request_token(scopes=[SCOPE], ttl="5m")

        secret = await vault.get_secret(SECRET_PATH)
        print(secret)

asyncio.run(main())

Non-blocking pattern

For production agents, don't block on approval. Instead, queue the task and resume after get_approval_status() returns an approved status.

Checking approval status

Approval decisions are made by humans in the dashboard. Agents receive an approval ID, then poll until the request is approved, denied, or expired.

Poll status

Shell

# Agents can poll an approval request by ID.
curl -X GET https://api.agentsecretstore.com/v1/approvals/agent/0b2e1c98-5ef3-4d2c-a7aa-45a91a18a201 \
  -H "Authorization: Bearer $ASS_AGENT_KEY"

Status payload

JSON

{
  "id": "0b2e1c98-5ef3-4d2c-a7aa-45a91a18a201",
  "tenant_id": "0e92b4ed-8e0e-4d83-b35a-a936fda75834",
  "agent_id": "0fd78c7f-53dd-4a87-8a0c-74b5a4c9a2db",
  "status": "pending",
  "requested_scopes": ["secrets:read:production/stripe/*"],
  "access_tier": "critical",
  "required_approvals": 1,
  "current_approvals": 0,
  "approvers": [],
  "reason": null,
  "denial_reason": null,
  "expires_at": "2026-06-01T11:00:00Z",
  "created_at": "2026-06-01T10:30:00Z",
  "resolved_at": null
}

Human action

Text

# Human approvers approve or deny requests in the dashboard:
https://agentsecretstore.com/dashboard/approvals