reply-triage - SKILL.md Agent Skill

name: reply-triage description: >- Surface high-value reply/engagement opportunities (X mentions, target tags, curated subreddit threads) and DRAFT suggested, value-first replies into the Notion approval queue for a human to post by hand. This skill NEVER posts, likes, follows, or replies via any API — automated engagement violates X's policy (fast suspension) and Reddit etiquette. TRIGGER on the scheduled engagement pass or on demand. SKIP any opportunity that can only be answered with a plug rather than genuine value.

reply-triage — human-gated engagement drafting

Purpose: replies are the highest-leverage reach mechanic, but automated replies get accounts suspended and read as spam. This skill does the finding and drafting; a human does the posting. It reads (read-only) a curated set of signals, drafts replies that add genuine value, and queues them in Notion with the link to reply to.

Load docs/social/brand-voice.md (incl. the Reddit-specific voice section), docs/social/social-runbook.md (the compliance invariant: no reply/like/follow endpoint is ever called), and docs/social/reddit-targets.md (the approved subreddit registry + per-sub rules — the source of truth for where and how to engage on Reddit).

Run this

Pull signals (read-only): recent X mentions of the founder/brand, posts on a curated target-tag list, and new threads in the approved subreddits from docs/social/reddit-targets.md (Tier-1 first). Reading is fine; do not call any write/engagement endpoint. Respect each sub's rule notes in the registry. For Reddit, run scripts/social/reddit-triage.js (@ascent/reddit; needs REDDIT_CLIENT_ID / REDDIT_CLIENT_SECRET / REDDIT_USER_AGENT, and a network path to Reddit) to get ranked candidate threads as JSON — it is read-only and never posts. Draft a reply for each high-value candidate.
Filter for value. Keep only opportunities where Ascent (or the founder) can add a genuinely useful answer — data, a concrete tip, a real perspective. Discard anything that could only be a plug. Quality bar is high; few is fine.
Draft the reply in the right register and platform etiquette. On Reddit: participant-first, product mentioned only if it directly answers, never the same text across subs. Cite any number to the Index.
Run brand-voice-guard on each draft.
Write to Notion at State = Review, Type = Reply, with the target URL to reply to in Body/Citations, Account, and for Reddit the Subreddit + its self-promo rule link. Mark clearly that a HUMAN posts these.

How to judge the result

Success: a short list of value-first reply drafts in Notion at Review, each with the exact target link, each passing brand-voice-guard, NONE posted automatically.
Verify the compliance invariant every run: confirm no engagement endpoint was called (this skill only reads + writes Notion). If a future change adds an auto-reply path, that is a compliance regression — stop and flag it (docs/social/social-runbook.md §1).
Prefer 2 excellent replies over 20 mediocre ones. Reply-guy volume without value is how accounts get muted.