redreplier

Track brand mentions across Reddit, Hacker News, X (Twitter), and Bluesky. RedReplier AI-scores each mention 0-100 for relevance so you focus on real leads instead of noise. Use this skill when you need to monitor keywords, triage leads, manage websites, configure alerts, or activate pending keywords.

intent

RedReplier ingests social mentions across four platforms and surfaces them ranked by relevance. this skill lets you manage monitored websites and keywords, list incoming mentions sorted by score or recency, approve or reject them for lead qualification, explain why the grader scored each mention, and set up email digests. the API is fully reversible except for two destructive actions (activating pending keywords with plan upgrades, deleting websites) which require explicit confirmation. use it when you want to catch product discussions before competitors do, find inbound leads hiding in comments, or build a searchable archive of what people say about your brand online.

inputs

environment

• REDREPLIER_API_KEY: bearer token, created at https://redreplier.com/signup under Settings → API Tokens. generate a dedicated, revocable token per agent (do not reuse across tools or humans). required.

external connection

• RedReplier SaaS at https://ai.redreplier.com/ai-app/api/v1. account determined by token, no account ID passed. ingests Reddit (posts and comments), Hacker News, X, and Bluesky in real-time. has rate limits (API docs site-specific; assume 10 req/sec, backoff if 429 returned).

context from user

• website domain/URL to monitor (optional if already added)
• keywords to track (phrases, product names, competitor names)
• mention filters (status: NEW/APPROVED/REJECTED, relevance bucket, source platform, date range, keyword subset)
• alert cadence (60/240/720/1440 minutes, clamped to plan minimum)

procedure

step 1: list monitored websites and keywords

input: REDREPLIER_API_KEY env var

call

GET /websites
header: Authorization: Bearer $REDREPLIER_API_KEY

output: JSON object { "websites": [{ "id", "domain", "url", "name", "description", "keywords": [{ "id", "value", "status" }] }] }. keyword status is one of PENDING, ACTIVE, DISABLED, SUSPENDED.

use: extract website IDs and keyword IDs for all downstream operations. if no websites exist, proceed to step 2.

step 2: add a website to monitor (optional)

input: website URL, optional name and description, initial keywords list

call

POST /websites
header: Authorization: Bearer $REDREPLIER_API_KEY
body: { "url": "https://example.com", "name": "Example", "keywords": ["example tool", "competitor name"] }

omit description to let RedReplier scrape and AI-generate one (used for relevance scoring context).

output: newly created website object with keyword IDs.

preview without creating: call POST /websites/analyze-description with { "url": "..." } first.

step 3: add keywords to a website

input: website ID (from step 1), list of keyword phrases

call

POST /websites/{WEBSITE_ID}/keywords
header: Authorization: Bearer $REDREPLIER_API_KEY
body: { "keywords": ["my product", "use case phrase"] }

keywords auto-activate up to plan limit; remainder stay PENDING.

output: list of added keywords with IDs and statuses.

related: edit with PATCH /keywords/{id} (metered against monthly allowance), disable with POST /keywords/{id}/disable, enable with POST /keywords/{id}/enable, delete with DELETE /keywords/{id} (PENDING only).

step 4: preview and activate pending keywords (paid action)

input: confirmation from user (yes/no) to upgrade plan if needed

step 4a: preview cost

GET /keywords/activate-pending/preview
header: Authorization: Bearer $REDREPLIER_API_KEY

output: { "pendingCount", "immediateCharge", "targetPlanName", "... }

show user immediateCharge and targetPlanName. if cost is zero, activation is free. if nonzero, confirm "yes" explicitly.

step 4b: activate (only after confirmation)

POST /keywords/activate-pending
header: Authorization: Bearer $REDREPLIER_API_KEY

output: list of newly activated keywords.

do not loop this call; activate once per user request.

step 5: list mentions (the leads)

input: filters (optional): website ID, statuses (NEW/APPROVED/REJECTED), relevance buckets (VERY_LOW/LOW/MEDIUM/HIGH/VERY_HIGH), sources (REDDIT_POST/REDDIT_COMMENT/TWITTER/BLUESKY/HACKERNEWS), keywords, date range (ISO 8601 from/to), sort (RELEVANCE or RECENT), limit (1-500), offset, includeLowRelevance (bool, default false)

call

GET /mentions?sort=RELEVANCE&limit=20&[filters]
header: Authorization: Bearer $REDREPLIER_API_KEY

defaults: excludes REJECTED, hides scores below 30. add &includeLowRelevance=true to see all.

example: this week's unreviewed high-relevance leads for one website

GET /mentions?websiteId=WEBSITE_ID&statuses=NEW&scoreBuckets=HIGH&scoreBuckets=VERY_HIGH&sort=RECENT

output: { "mentions": [...], "total", "limit", "offset }. each mention has relevanceScore, relevanceReason, tags, keyword, title, contentText, url, author, subreddit (Reddit only, null for X/Bluesky/HN), source, status.

count only (no data)

GET /mentions/count?[filters]

step 6: explain a mention's relevance score

input: mention ID (from step 5)

call

POST /mentions/{MENTION_ID}/explain
header: Authorization: Bearer $REDREPLIER_API_KEY

output: mention object with relevanceReason and tags populated (generated on-demand if missing).

use when score is borderline or reasoning is unclear before triaging.

step 7: triage a mention (approve, reject, or reset)

input: mention ID, new status (APPROVED, REJECTED, or NEW)

call

PATCH /mentions/{MENTION_ID}/status
header: Authorization: Bearer $REDREPLIER_API_KEY
body: { "status": "APPROVED" }

output: updated mention object.

meanings: APPROVED = real lead (kept), REJECTED = noise (hidden from default lists), NEW = back to inbox. all reversible.

step 8: configure email alerts

input: enabled (bool), cadenceMinutes (60, 240, 720, or 1440)

step 8a: read current settings

GET /alert-settings
header: Authorization: Bearer $REDREPLIER_API_KEY

output: { "enabled", "cadenceMinutes", "minIntervalMinutes" (plan limit)}.

step 8b: set new cadence

PUT /alert-settings
header: Authorization: Bearer $REDREPLIER_API_KEY
body: { "enabled": true, "cadenceMinutes": 240 }

cadenceMinutes is clamped up to plan's minIntervalMinutes.

output: resolved settings (confirms actual cadence applied).

decision points

if user wants to add keywords but you must upgrade the plan (step 4)

• call preview first, show cost and target plan name.
• if cost is zero, proceed to activate.
• if cost is nonzero, ask user "activate and upgrade to [PLAN]? this will charge $[AMOUNT]."
• only activate after explicit user yes.
• never activate in a loop.

if user wants to delete a website (destructive)

• confirm with user first: "delete [DOMAIN_NAME]? this stops all monitoring."
• do not delete based on ID alone.
• only proceed after explicit user yes.

if a keyword is SUSPENDED

• do not try to force it back to ACTIVE by looping edits.
• tell user: "this keyword was auto-judged too noisy. edit it to fix the wording, then resubmit for re-grading (free)."
• offer to edit wording or propose a narrower synonym.

if user wants to edit keywords frequently

• warn: "edits count against your monthly allowance. prefer adding new keywords or disabling old ones."
• check GET /keywords/change-usage to show remaining budget.
• only edit for wording fixes, not cosmetic tweaks.

if mention score is below 30 or user asks for "all mentions"

• add &includeLowRelevance=true to the request.
• warn user: "low-relevance results include heavy noise (VERY_LOW and LOW buckets)."

if user pulls unreviewed mentions (NEW status) and wants to triage at scale

• list with statuses=NEW&sort=RELEVANCE to show best first.
• for each mention, show source (REDDIT_POST, TWITTER, BLUESKY, HACKERNEWS), subreddit (Reddit only), score, and reason.
• batch-ask user which to APPROVE before making calls.
• do not approve based on score alone, verify relevanceReason and contentText match the user's product/use case.

if API returns 429 (rate limit)

• backoff with exponential delay (2s, 4s, 8s, ...).
• do not retry in a tight loop.
• inform user: "api rate-limited, retrying..."

if token is invalid or expired

• return: "REDREPLIER_API_KEY invalid or revoked. regenerate at https://redreplier.com/settings."

output contract

successful list of websites

• JSON array of website objects with nested keywords, or empty array if none exist.

successful mention triage

• HTTP 200 with updated mention object, status field now reflects the new value (APPROVED/REJECTED/NEW).

successful keyword activation

• HTTP 200 with list of newly activated keyword objects, status changed from PENDING to ACTIVE.

successful alert configuration

• HTTP 200 with alert settings object confirming enabled state and actual cadenceMinutes applied (may differ from request if clamped to plan).

file/location: no files written to disk. all state lives in RedReplier account (read via step 1 list).

edge cases

• empty mentions list: user has no matches for that filter combination. inform user clearly (not an error).
• pending keywords, no activation: stay PENDING indefinitely until user calls step 4 (no auto-activation).
• mention with missing relevanceReason: call explain (step 6) to populate on-demand.
• network timeout (>30s): retry once, then error with "redeplier api unreachable, try again in 1 minute."
• token with no websites: start from step 2.
• plan downgrade after activation: previously activated keywords become disabled (not deleted). user must re-enable or upgrade to use them again.

outcome signal

user knows it worked when

• GET /websites returns the list and user sees their domains.
• after adding keywords, GET /websites shows them with correct IDs and status (PENDING or ACTIVE).
• after POST /keywords/activate-pending, those keywords move from PENDING to ACTIVE in the list.
• GET /mentions returns a list with scores, reasons, and sources. user can see at least one mention (or zero if no matches, which is valid).
• after PATCH /mentions/{id}/status, that mention's status changes in subsequent GET /mentions calls (REJECTED mentions disappear from default lists).
• after PUT /alert-settings, user receives an email digest at the configured cadence (first email arrives within cadenceMinutes of enable).
• user can navigate back to https://redreplier.com/dashboard and see the same websites, keywords, mentions, and alert settings in the UI.