| name | agent-dispatch |
| description | Use `al dispatch` for focused headless second-agent work — another model, tool set, skill, or review perspective. Trigger when the user asks to run al dispatch, use a subagent or second agent, get a fresh-context review, or invoke a target agent with a skill. Do not use for ordinary shell commands, local test runs, web search, browser automation, or multi-agent orchestration. |
| compatibility | Requires the Agent Layer CLI (`al`) from the project environment and at least one configured target for actual al dispatch runs. |
| allowed-tools | Bash(al:*) Bash(cat:*) |
al dispatch
Use al dispatch as Agent Layer's headless second-agent control surface. This
skill provides routing, safety, and workflow rules; installed CLI help provides
command syntax and runtime contract details.
Defaults
- Keep the target prompt focused on the second agent's job.
- Create no durable artifact unless the task needs one.
Required artifacts
- No artifact is required for ordinary
al dispatch runs.
- If preparing prompt files or saving target output, put agent-only files under
.agent-layer/tmp/.
- Use
.agent-layer/tmp/agent-dispatch.<run-id>.<type>.md or
.agent-layer/tmp/agent-dispatch.<run-id>.<type>.json for scratch prompt,
output, or options files.
- Report every artifact path created.
Global constraints
- Run
al dispatch --help before the first al dispatch command in a session.
- Run relevant subcommand help before using non-obvious subcommands or flags.
- Treat installed CLI help as the source of truth for commands, arguments,
flags, defaults, and runtime contract details.
- If
al, a target CLI, authentication, source skill, or generated skill
projection is missing, stop and report the missing requirement. Do not
install, authenticate, sync, or change configuration unless the user asked
for setup or repair.
Command routing
Use al dispatch when the task benefits from a separate headless agent: a
different model, a target-specific tool set, a target skill, or a fresh review
perspective. Do not use it for work the current agent can perform directly with
local shell commands.
Workflow
- Inspect
al dispatch --help.
- Follow live help to inspect target metadata before a real run.
- Choose the smallest target path that matches the user's intent.
- Verify any requested skill exists in the current project's Agent Layer skill
source, or let
al dispatch validation fail clearly. Do not invent a skill name.
- Send the focused prompt to one target. Omit unrelated conversation
history and information that could bias the target.
- If a prompt file is needed, write it under
.agent-layer/tmp/.
Guardrails
- Do not use
al dispatch for ordinary local shell work, tests, web retrieval,
browser automation, or API-only tasks.
- Do not use
al dispatch to bypass the caller's restrictions, approvals, or
sandbox expectations.
- Do not retry failed target launches by guessing flags, targets, models, or
reasoning-effort values. Re-check help and options, then report the mismatch
if it remains unresolved.
Definition of done
- Live help was checked for each command shape used.
- Current target metadata was inspected before any real run.
- Target selection and skill delivery if used were intentional and reported.
- Prompt and output artifacts were stored only under
.agent-layer/tmp/ and
every created artifact path was reported.
- Missing targets, authentication, skill projection, or unsupported options were
reported explicitly.
- Any side effects from target work were inspected and verified through the
current task's normal checks.
Final handoff
State the target selected, whether a skill was invoked, how the prompt was
provided, the useful result from the target, any artifacts created, and any
verification or al dispatch limits that could not be resolved.