← Web App

Agent editor

web app · wireframes · agents shell loading

Create and edit a single agent: prompt, tools, schedule — plus a manual run right here with a run log. One screen for both creation (/agents/new) and editing an existing agent.

Configuration and runs → Agent Engine
achilles.local/agents/12 /agents/:id · agent owner Creation — /agents/new.
AAchilles DS
Release digest active Enabled 1
Basics
Name
Release digest
Description
A summary of Ajax project task statuses and open blockers.
optional
Prompt 3
On every run, put together a short digest of the Ajax project: what closed in the period, which tasks are in progress and which blockers are open. Answer as a list, with a source link for each item.
Run settings
Model
Claude Sonnet
from the list allowed by the administrator
Internal tools 4
search graph sql
core · under your permissions · read-only
External tools 10
web_search fetch_url
from what the administrator allowed · read-only · click to toggle; unchecking doesn't block the run
Connect an MCP server v2
Schedule 5
Interval
every 6 hours
mode + parameter; the parameter adapts to the mode — preview below. Manual · interval · calendar.
Narrowing data scope — v2; for now the agent sees your full access.

Run log 6
7 tasks closed in the period; 3 in progress (AJX-218, AJX-224, AJX-231). Open blocker: the architecture review for AJX-218 is waiting on the release manager. Deadline pushed to June 30.
Yesterday 14:00 scheduled · succeeded · 11 s · 31k tokens
2 days ago failed · iteration cap 7

  1. 1 Owner switch — "enabled" keeps the agent on its schedule and available for manual runs, "disabled" stops it entirely. The summary status sits in the header.
  2. 2 "Run now" — an ad-hoc run with the current settings; respects the switch, the admin lock and the budget. One active run per agent: another on top of a running one is skipped. Agent Engine.
  3. 3 The owner's prompt — the task, output format, constraints. It layers on top of the platform layer (safety + organization rules) and can't override it. Agent Engine.
  4. 4 The core of three internal tools — search, graph, SQL — is fixed and always runs under the owner's identity: what's closed to the employee is closed to the agent, and writing to the graph is unavailable. Agent Engine.
  5. 5 Schedule — a mode plus a parameter for it (modes previewed above): "manual" — no parameter, interval — an "every N hours" preset, calendar — a "daily / weekly at HH:MM" slot, resolved against the owner's timezone. Raw cron and event triggers — v2. Agent Engine.
  6. 6 Run log — time, trigger, outcome, duration and token spend; a row expands into the agent's full output.
  7. 7 Run outcome: succeeded · failed (error / iteration cap) · skipped (weekly limit / already running). The state machine — Agent Engine.
  8. 8 The admin lock is sticky: while it's set, the switch and runs are unavailable to the owner. Agent Engine.
  9. 9 The model was removed from the list by an administrator — the agent is stopped until the owner selects another from the current list. The previous choice doesn't fall back to the default. Agent Engine.
  10. 10 External tools — a choice from the read-only presets the administrator allowed (web_search, fetch_url); unchecking one doesn't block the run, the agent carries on with the KS core. Connecting MCP servers — v2. Admin Panel.