EchoNet.ing
PHASE B / CAPABILITY ATLAS

Tools & Permissions

A capability can be installed, enabled, and still not authorized for the current actor, target, or action.

Availabletechnically installed
Enabledexposed by policy
Authorizedallowed for this scope
Invokedexplicit input
Verifiedread-back or runtime proof
OPERATING MODEL

Tools & Permissions

Tool safety depends on three separate decisions: whether a capability exists, whether profile policy exposes it, and whether the current task authorizes its use. Read tools still need privacy limits. Write and external-action tools need tighter scope, approval, and read-back.

Inventory capabilities

Record purpose, side effects, requirements, and data exposure.

Set profile policy

Enable only families needed for that profile’s role.

Classify each action

Separate safe reads, reversible writes, external writes, destructive actions, and prohibited operations.

Require explicit targets

Bind each invocation to named inputs and destinations.

Verify effects

Read the exact target after writes and preserve evidence without secrets.

PRACTICAL REFERENCE

18 enabled public-safe families

webSearch and extract public web sources
browserRender pages and exercise interactions
fileRead, search, write, and patch scoped files
visionInspect screenshots and images
X searchSearch public posts and profiles
session searchRecover relevant conversation history
memoryRead or update compact stable facts
context engineAssemble task-relevant context
code executionRun bounded data-processing logic
skillsLoad and maintain reusable procedures
todoTrack multi-step work
clarifyRequest decisions when ambiguity matters
image generationCreate visual assets
TTSCreate spoken audio
terminalRun commands and managed processes
computer useInteract with approved local applications
delegationAssign isolated bounded subtasks
cronRun durable scheduled jobs
FAILURE MODES

What breaks—and how to recover.

Availability mistaken for permission

Check profile policy and task authorization separately.

Decorative tool blocks controls

Use pointer-event and hit-testing checks for browser layers.

Write succeeds but task fails

Verify the target state, not only the tool response.

Copyable implementation prompt

Create a least-privilege tool policy for [PROFILE]. Group tools into Read, Reason, Create, Act, and Coordinate. For each define availability, enablement, authorization, data boundary, approval level, side effects, rollback, and proof.
Public boundary

Use placeholders for paths, accounts, endpoints, identities, and credentials. Never publish secret values, private records, or a live control surface.

Verification gate

  • All 18 relevant families are classified
  • External and destructive effects require approval
  • Secrets are excluded from prompts and logs
  • Writes have exact read-back checks