Reply Engine

What it does

When your agent uses it

• The user types /reply-engine
• Says draft the suggested replies
• Run the reply pass
• Build the reply sheet

Example prompts

• “/reply-engine”
• “draft the suggested replies”
• “run the reply pass”
• “/reply-engine”

Requirements

• Python 3

Workflow steps

8 steps, taken from the first numbered list in SKILL.md.

1. 1Answer the actual ask. One or two clipped sentences; the first responds to their literal question.
2. 2Concrete nouns from the buyer's world. Their tools, their workflows, their numbers, never category abstractions.
3. 3Zero product mentions by default. Across a whole batch, a mention is the rare exception, and naming the product triggers the disclosure…
4. 4End with a question back or a first step they can take today.
5. 5Numbers stated plainly. "Free tier plus two hours of setup beats a paid seat nobody opens." No hedging.
6. 6Match the thread's language and register. A Spanish thread gets a Spanish reply.
7. 7No links, no DMs, no CTAs, no emojis, no "great question".
8. 8If it can't fit in 18 words, it's two thoughts. Cut one.

What it can do on your machine

Read from SKILL.md and the folder at commit 072c185. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Shell commands in SKILL.md call:

  • python3

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Links to these hosts (documentation or services it may open):

  • github.com

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Reply Engine loads about 1.8k tokens when it runs. Until then it costs about 166 tokens; SKILL.md has 728 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~166

When it runs · the whole SKILL.md, loaded when a task matches

~1.8k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from shawnla90/gtm-coding-agent at commit 072c185, republished under its MIT licence (© shawnla90). 728 words, ~1,812 tokens.

name

reply-engine

description

Batch-draft one suggested Reddit reply template per classified opportunity, hard-gated. Wraps the reddit-buyer-signals starter's replies.py — scaffold builds suggested_replies.json from the classified ops, the agent writes each ≤18-word value-first template, check enforces the word cap and the slop gate, sheet renders a rules-pinned Suggested Replies tab with GO/REVIEW/NO-REPLY on every row, and angles feeds the daily digest. Every reply is a draft the human edits before posting; NO-REPLY rows are logged, never answered. Use when the user types "/reply-engine" or says "draft the suggested replies", "run the reply pass", "build the reply sheet".

version

1.0.0

cd starters/reddit-buyer-signals
python3 replies.py scaffold --ops data/ops_classified.json --out data/suggested_replies.json
# the agent writes every empty reply slot, then:
python3 replies.py check data/suggested_replies.json --ops data/ops_classified.json
python3 replies.py sheet --ops data/ops_classified.json --replies data/suggested_replies.json --sheet-id <id>
python3 replies.py angles --ops data/ops_classified.json --replies data/suggested_replies.json --out data/engage_angles.json

printf '%s' "<reply>" | wc -w

{
  "rules": "Every reply is a draft template. Nothing posts automatically. All replies go through the account owner personally before posting. Disclose affiliation whenever the product comes up. Never DM. Check subreddit self-promo rules and thread age before posting. Gate: GO = timely, reply when ready. REVIEW = check thread age + sub rules first. NO-REPLY = log as intel, do not post.",
  "replies": {
    "op_101": "Twelve people will outgrow both boards the same way. What breaks first for you: reporting or handoffs?"
  },
  "gate_overrides": {
    "op_207": ["NO-REPLY", "Founder self-promo. Flag for partnership outreach instead."]
  }
}

© shawnla90, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in skills/reply-engine of shawnla90/gtm-coding-agent.

Open the folder on GitHubat commit 072c185