Skip to content
Back to skills

Dispatch Packet

ASecurity

Generate a dispatch packet — a self-contained subagent packet / implementer prompt that lets a weaker executor equal a stronger one on bounded work. Use when asked to generate a subagent packet, dispatch an implementer, turn issue N into an agent prompt, or write a delegation prompt for a sonnet/opus executor from an issue or task description. Never auto-dispatches — a human/orchestrator reviews the generated packet before any agent runs it.

  • 4 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 6, 2026
ai-agentsgoshellgitapisecurity

Works with

  • api

Security analysis

A100/100

Pro scans all 4 files and shows the line behind each finding

Scanned September 6, 2026

npx -y skills add fakoli/fakoli-plugins --skill dispatch-packet --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dispatch Packet?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dispatch Packet
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/fakoli-dispatch-packet/badge)](https://www.skillsdirectory.com/skills/fakoli-dispatch-packet)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dispatch-packet
description: Generate a dispatch packet — a self-contained subagent packet / implementer prompt that lets a weaker executor equal a stronger one on bounded work. Use when asked to generate a subagent packet, dispatch an implementer, turn issue N into an agent prompt, or write a delegation prompt for a sonnet/opus executor from an issue or task description. Never auto-dispatches — a human/orchestrator reviews the generated packet before any agent runs it.
user-invocable: true
---

# Dispatch Packet

Turn an issue (`gh issue view N --json title,body,comments`) or a task
description into a dispatch packet: a delegation prompt with every decision
closed, every anchor grep-verified, and a mirror-PR to copy the shape of.
Freehand subagent prompts stalled 2/7 in practice; packets with this
discipline shipped 8/8. Template: `references/packet-template.md` (format
`v2`) · rationale + model-tier rubric: `references/dispatch-discipline.md`
· worked example: `references/example-packet-377.md`.

## 1. Read context first

Read the source issue/task, its linked playbook or design doc, and any
"mirror this PR" precedent before drafting anything. Playbook first, then
persona/spec docs, then the precedent PRs.

## 2. Restate the decided design — never resolve an open one

Enumerate the decisions the source already closed under `## Decided
design`, and record `MODEL TIER: <sonnet|opus>` per the rubric in
`dispatch-discipline.md`. An `OPEN-DECISION` or unresolved tension goes
under `## Open decisions (BLOCKING)` instead, never Decided design — that
marks the header `STATUS: BLOCKED` and returns the packet to the
requester, not an executor.

## 3. Anchor every file:line reference

For every anchor you plan to cite (`file.py:123`, a function name to grep),
run the grep yourself against the target repo/worktree and embed the
evidence line in the packet. A stale anchor sent an agent chasing an API
signature that had moved. If you cannot run the grep (wrong repo checked
out), say so in the packet rather than presenting a guess as fact.

## 4. Mirror a merged PR

Point at the closest merged PR that shipped the same shape of change and
list the exact edits it made, file:line — the single highest success
predictor observed.

## 5. Boundaries

List files/directories the executor must NOT touch and why (who owns them,
what breaks). State explicitly: no `git commit`, `git push`, `git checkout`,
or branch creation — the orchestrator owns git. Two agents may share one
worktree only if their file sets are disjoint; otherwise use
`isolation: "worktree"`.

## 6. Tests

Name the test file to create, the fixture/helper to copy patterns from
(injectable-runner pattern: fake the subprocess/`_run` seam, don't shell
out in tests), the specific cases, and require tests to fail against a
broken implementation.

## 7. Verify

Exact commands, in order. `git add -A` MUST run before any tooling that
reads `git ls-files` (a stale index passes locally, fails in CI). Require
verbatim command tails in the report, not a summary.

## 8. Anti-stall clause

Every packet header carries the two-line anti-stall block verbatim from
`packet-template.md` — start deadline plus stop-and-report. 2 of 7
freehand-dispatched agents stalled in open-ended reading.

## 9. `[operator]` markers

Anything needing live hosts, credentials, or SSH gets an `[operator]`
section — exact commands produced, not run. State it even when empty.

## 10. Report back

Require: files changed (path + line count), chokepoints chosen and why,
verify-command tails verbatim, and any out-of-scope observation — reported,
not silently fixed. **If the dispatched task is itself a review**, the
packet must additionally require a REPRO per finding, severity ranking, an
explicit SHIP / DO-NOT-SHIP verdict, and `SUSPECTED` on anything the
reviewer could not reproduce (`references/dispatch-discipline.md`).

## Refuse rather than guess

If the source material lacks decisions, emit the skeleton from
`packet-template.md` with the gaps under `## Open decisions (BLOCKING)`
and `STATUS: BLOCKED` in the header — return it to the requester, not an
executor. Do not fabricate a design; link oversized detail to reference
docs instead of inlining it, EXCEPT hardened/security code, always
inlined.

## No auto-dispatch

This skill only generates the packet — dispatching it is a separate,
human-reviewed step; reviewing the packet itself is the quality gate.

Files in this skill

  • SKILL.md4.4 KB
  • references/dispatch-discipline.md3.1 KB
  • references/example-packet-377.md12.8 KB
  • references/packet-template.md5.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.