agentleFS
synaptiaisynaptiai/agent-capability-standard/skills/send/SKILL.md

send

Skill6 starsChanged 13 days agoApache-2.0View on GitHub

Emit a message or event to an external system with policy enforcement and approval gates. Use when publishing messages, calling APIs, sending notifications, or triggering external workflows. REQUIRES EXPLICIT APPROVAL.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

24 rules. 0 are common, 22 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 22

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    All policy constraints verified before sending

    line 57
  2. Distinctive:

    Check the approval is valid and recent

    line 78
  3. Distinctive:

    Validate destination: Ensure target is permitted

    line 87
  4. Distinctive:

    Check destination against allowlist if defined

    line 88
  5. Distinctive:

    Verify destination format matches protocol

    line 89
  6. Distinctive:

    Check rate limits have not been exceeded

    line 93
  7. Distinctive:

    Verify message size within bounds

    line 94
  8. Distinctive:

    Add required headers/metadata

    line 100
  9. Distinctive:

    Use appropriate protocol handler

    line 105
  10. Distinctive:

    Check response status/acknowledgment

    line 111
  11. Distinctive:

    When it was sent

    line 118
  12. Distinctive:

    Verification tools: Read (for logs), Bash (for health checks)

    line 280
  13. Distinctive:

    NEVER send without explicit user approval (CRITICAL)

    line 290
  14. Distinctive:

    NEVER send without a valid checkpoint reference

    line 291
  15. Distinctive:

    ALWAYS validate destination against policy before sending

    line 292
  16. Distinctive:

    ALWAYS log what was sent for audit (sanitize sensitive data)

    line 293
  17. Distinctive:

    DO NOT retry indefinitely - respect retry limits

    line 294
  18. Distinctive:

    If approval token is missing, REQUEST approval before proceeding

    line 296
  19. Distinctive:

    Never send credentials, API keys, or secrets in payloads

    line 297
  20. Distinctive:

    Before any send operation, the user must explicitly approve:

    line 300
  21. Very specific:

    NEVER send without constrain check (policy bypass)

    line 321
  22. Very specific:

    NEVER send without checkpoint (no state recovery)

    line 322
  23. Distinctive:

    NEVER send without explicit approval (unauthorized action)

    line 323
  24. Distinctive:

    NEVER retry failed sends indefinitely (resource exhaustion)

    line 324

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Intent

Send a message, event, or payload to an external destination with full policy enforcement, approval gates, and audit trail. This capability handles external side effects that cannot be undone.

CRITICAL: This capability requires explicit user approval before execution.

Success criteria:

  • Message delivered to destination successfully
  • All policy constraints verified before sending
  • Approval gate passed (explicit user confirmation)
  • Complete audit trail of what was sent
  • Delivery confirmation obtained where possible

Compatible schemas:

  • schemas/output_schema.yaml

Inputs

ParameterRequiredTypeDescription
destinationYesstringTarget endpoint, queue, API, or address
payloadYesanyData to send
protocolNostringCommunication protocol (http, grpc, smtp, amqp, etc.)

Inputs not listed here are outside the capability's contract: a workflow step that binds them gets W604 (STANDARD v1.1 §5.5).

Procedure

  1. Verify approval: Confirm explicit user approval exists
    • Check the approval is valid and recent
    • BLOCK send if approval is missing or expired
    • Log approval verification for audit
  2. Verify checkpoint: Confirm recovery point exists
    • Validate that a checkpoint was created before this send
    • BLOCK send if checkpoint is missing
    • External sends are irreversible - checkpoint is for state recovery
  3. Validate destination: Ensure target is permitted
    • Check destination against allowlist if defined
    • Verify destination format matches protocol
    • Block sends to disallowed destinations
  4. Apply constraints: Enforce policy limits
    • Check rate limits have not been exceeded
    • Verify message size within bounds
    • Validate message format against schema
    • Apply constrain capability rules
  5. Prepare payload: Format message for transmission
    • Serialize according to protocol requirements
    • Add required headers/metadata
    • Generate message ID for tracking
    • Record payload summary (not full content if sensitive)
  6. Execute send: Transmit to destination
    • Use appropriate protocol handler
    • Capture response or confirmation
    • Handle timeouts gracefully
    • Retry according to policy if configured
  7. Verify delivery: Confirm successful transmission
    • Check response status/acknowledgment
    • Log delivery confirmation
    • Capture any error responses
  8. Audit trail: Record complete send operation
    • What was sent (summary, not sensitive content)
    • Where it was sent
    • When it was sent
    • Delivery status
    • Approval reference

Output Contract

Return a structured object:

sent: boolean  # true when the transport accepted the message (delivery.status sent or queued)
response: any  # Recipient's response or delivery receipt, if any
timestamp: string  # ISO 8601 date-time of send
message:
  destination: string  # Where message was sent
  payload_type: string  # Type of content sent
  payload_summary: string  # Non-sensitive summary of payload
  message_id: string | null  # Tracking ID if available
delivery:
  status: sent | queued | failed  # Delivery outcome
  retry_count: integer  # Number of retry attempts
  response_code: integer | null  # HTTP status or protocol response
  error: string | null  # Error message if failed
constraints_applied:
  - constraint: string  # What constraint was checked
    enforced: boolean  # Whether it passed
    details: string | null  # Additional context
approval:
  token: string  # Approval token used
  timestamp: string  # When approval was given
  scope: string  # What was approved
confidence: 0..1  # Confidence in successful delivery
evidence_anchors: ["log:timestamp", "api:response"]
assumptions: []  # Explicit assumptions

Field Definitions

FieldTypeDescription
sentbooleanWhether the message was sent (accepted by the transport)
responseanyRecipient's response or delivery receipt, or null
timestampstringISO 8601 date-time of send
messageobjectDetails of what was sent (destination, payload type, non-sensitive summary, message ID)
deliveryobjectOutcome of send operation
constraints_appliedarrayPolicy checks performed
approvalobjectApproval gate information
confidencenumber0.0-1.0 based on delivery confirmation
evidence_anchorsarrayLogs, responses, confirmations
assumptionsarrayExplicit assumptions (e.g., "network stable")

Examples

Example 1: API Webhook Notification

Input:

destination: "https://api.partner.com/webhooks/order-updates"
payload:
  event: "order.completed"
  order_id: "ORD-12345"
  timestamp: "2024-01-15T10:30:00Z"
  customer_notified: true
protocol: "http"

Approval and the checkpoint come from the workflow (an approved step after a checkpoint step), not from inputs. A preceding constrain step enforces the policy: at most 100 sends per hour, to api.partner.com only.

Output:

sent: true
response:
  received: true
timestamp: "2024-01-15T10:30:15Z"
message:
  destination: "https://api.partner.com/webhooks/order-updates"
  payload_type: "order.completed event"
  payload_summary: "Order completion notification for ORD-12345"
  message_id: "msg-abc123"
delivery:
  status: sent
  retry_count: 0
  response_code: 200
  error: null
constraints_applied:
  - constraint: "rate_limit: 100/hour"
    enforced: true
    details: "Current rate: 45/hour"
  - constraint: "allowed_domains"
    enforced: true
    details: "api.partner.com is in allowlist"
approval:
  token: "approve-send-2024-01-15-001"
  timestamp: "2024-01-15T10:30:00Z"
  scope: "Send order notification to partner API"
confidence: 0.95
evidence_anchors:
  - "log:2024-01-15T10:30:15Z:send-success"
  - "api:response:200-OK"
assumptions:
  - "Partner API is available and processing webhooks"

Evidence pattern: HTTP response code and body confirmation, timestamped log entry.


Example 2: Email Notification (with failure)

Input:

destination: "smtp://mail.company.com:587"
payload:
  to: "user@example.com"
  subject: "Action Required: Review pending"
  body: "You have 3 items awaiting review..."
protocol: "smtp"

Output:

sent: false
response: null
timestamp: "2024-01-15T11:00:05Z"
message:
  destination: "smtp://mail.company.com:587"
  payload_type: "email"
  payload_summary: "Email to user@example.com regarding pending review"
  message_id: null
delivery:
  status: failed
  retry_count: 3
  response_code: null
  error: "Connection timeout after 30s - SMTP server unreachable"
constraints_applied:
  - constraint: "recipient_validation"
    enforced: true
    details: "Email format valid"
approval:
  token: "approve-email-2024-01-15-002"
  timestamp: "2024-01-15T11:00:00Z"
  scope: "Send review notification email"
confidence: 0.0
evidence_anchors:
  - "log:2024-01-15T11:00:05Z:smtp-timeout"
  - "error:connection-refused"
assumptions:
  - "SMTP server should be available on port 587"
next_actions:
  - "Verify SMTP server status"
  - "Check network connectivity"
  - "Retry with backup mail server if available"

Verification

  • Approval token is valid and not expired
  • Checkpoint reference exists
  • Destination passes policy validation
  • All constraints were checked before send
  • Delivery status is accurately reported
  • Audit trail is complete

Verification tools: Read (for logs), Bash (for health checks)

Safety Constraints

  • mutation: true
  • requires_checkpoint: true
  • requires_approval: true
  • risk: high

Capability-specific rules:

  • NEVER send without explicit user approval (CRITICAL)
  • NEVER send without a valid checkpoint reference
  • ALWAYS validate destination against policy before sending
  • ALWAYS log what was sent for audit (sanitize sensitive data)
  • DO NOT retry indefinitely - respect retry limits
  • BLOCK sends to destinations not in allowlist when list is defined
  • If approval token is missing, REQUEST approval before proceeding
  • Never send credentials, API keys, or secrets in payloads

APPROVAL GATE REQUIREMENT: Before any send operation, the user must explicitly approve:

  1. What will be sent (summary)
  2. Where it will be sent (destination)
  3. Why it is being sent (context)

The approval must be captured with a token that is recorded in this operation's audit trail.

Composition Patterns

Commonly follows:

  • constrain - Apply policy limits before sending (REQUIRED)
  • checkpoint - Create recovery point before external send (REQUIRED)
  • plan - Sends are typically part of a larger plan
  • transform - Format data before sending

Commonly precedes:

  • audit - Record what was sent for compliance
  • verify - Confirm delivery where possible
  • receive - May expect response from destination

Anti-patterns:

  • NEVER send without constrain check (policy bypass)
  • NEVER send without checkpoint (no state recovery)
  • NEVER send without explicit approval (unauthorized action)
  • NEVER retry failed sends indefinitely (resource exhaustion)

Workflow references:

  • See reference/composition_patterns.md#anti-patterns for send safety rules
  • See reference/composition_patterns.md#digital-twin-sync-loop for send in sync context

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.