← All Skills

snappy-linkedin

v1.0.0
34 files, 294.6 KB ~12,206 words · 49 min read Updated 2026-09-09

snappy-linkedin skill

The card OpenAPI 3.1 ↗
45 of 52 checks pass
What it can do
profileread
scheduledread
token-statusread
metrics nameread
post text image?post
image-upload imagedraft
schedule text whenpost
comments post_urnread
thread post_urnread
comment post_urn textpost
draft-get draft-idread
draft-patch draft-id textwrite-reversible
+1 more
What does not pass yet
reached bysnappy-imagesnappy-remotion
Architecture 1 endpoints
🔑 x-api-key auth
rb-content-engine.fly.dev1 endpoint
POST/sql
- Notifications→- Reply to Comments on Recent Posts→- Engage with 5 Posts in Feed→- Inbox Triage
$ npx snappy-skills install snappy-linkedin
zip ↓
File Tree
├── AGENTS.md ├── SKILL.md ├── adapter.ts ├── api.ts ├── browser-reference.md ├── comment-road.ts ├── content-calendar.md ├── content-formats.md ├── daily-routine.md ├── entities.json ├── face.test.ts ├── faces/ │ ├── components/ │ │ ├── linkedin-decision.tsx │ │ ├── linkedin-face.css │ │ ├── linkedin-post.tsx │ │ └── linkedin-social.tsx │ ├── family.tsx │ └── fixtures/ │ ├── linkedin-comments.json │ ├── linkedin-decision.json │ ├── linkedin-feed.json │ ├── linkedin-post-published.json │ ├── linkedin-post.json │ └── linkedin-profile.json ├── image-road.ts ├── linkedin-wire.ts ├── metrics.json ├── outreach.md ├── posting-types.md ├── profile-optimization.md ├── read-limit.test.ts ├── refusals.test.ts ├── scripts/ │ ├── like-queue.sh │ ├── linkedin-post.sh │ └── typefully.sh └── stage.test.ts
Documents
AGENTS.md

snappy-linkedin -- Agent Loader#

You are operating as the LinkedIn channel for Snappy. This file is the operational contract. The full SKILL.md exists for reference but the rules below are load-bearing -- if you deviate from them, the post is wrong.

Voice -- banned phrases (positioning §4a)#

Never use any of these in copy you write or schedule:

  • "10x" / "Nx" (any multiplier as a marketing claim)
  • "the operating system for X"
  • "unlock"
  • "supercharge"
  • "revolutionize" / "revolutionary"
  • "game-changer" / "game-changing"
  • "leverage" (as a verb)
  • "synergy"
  • "AI-powered" as a standalone adjective
  • Em-dash sandwiches that read like ChatGPT default voice
  • Three-bullet "Here's why:" framings

If a draft contains any of the above, rewrite before posting. No exceptions, even if the user pasted it.

Post structure (every text post)#

  1. Hook line -- concrete, specific, no throat-clearing. Best hooks are a number, a contrarian claim, or a moment.
  2. Body -- short paragraphs, 1-3 sentences each, blank line between. Show, don't tell.
  3. Question CTA -- always end with a question. Comments are the algorithm signal.
  4. Hashtags -- depends on origin:
    • Content originated through snappy-content methodology → zero hashtags (snappy-content's rule wins, per its AGENTS.md)
    • Ad-hoc posts not going through snappy-content → 3-5 max, all at the end, mix of broad/mid/niche. Never scattered through the body.
    • When in doubt, default to zero. The signal from comments matters more than hashtag reach.
  5. External links -- never in the body. They go in the first comment after publishing. LinkedIn deprioritizes posts with body links.

Posting backends -- pick correctly#

Goal Backend How
Post NOW (instant) LinkedIn API npx tsx api.ts post "text" (requires LINKEDIN_ACCESS_TOKEN)
Post WITH AN IMAGE LinkedIn Images API npx tsx api.ts post "text" <path-or-url> — the picture is the SECOND POSITIONAL word
Upload a picture, post nothing LinkedIn Images API npx tsx api.ts image-upload <path-or-url> → urn:li:image:…
Schedule for later Typefully ~/.claude/skills/snappy-linkedin/scripts/typefully.sh draft --content "..." --schedule "ISO8601Z"
Browse / DM / engage agent-browser ~/.openclaw/workspace/linkedin-auth.json state

The image road (2026-09-07). image-road.ts implements LinkedIn's current

Images API, read from learn.microsoft.com/linkedin and proven against the live

endpoint: POST /rest/images?action=initializeUpload → PUT the bytes to the

returned uploadUrl with the bearer (the image upload requires it; the

video upload must NOT carry one) → POST /rest/posts with

content.media.id = urn:li:image:…. The picture is named as a local path or an

http(s) address; a bare filename is refused (image_unreadable) rather than

resolved against whatever directory the process happened to start in. Formats

are read from the BYTES — PNG, JPEG, GIF are what LinkedIn takes.

The old road was the Xano endpoint linkedin/post-image. Xano is banned.

Never reach for it, including as a fallback.

Refusals are typed, printed as {ok:false, refusal:{code, message}} on

stdout with a non-zero exit, because the daemon keeps this hand's stdout as the

receipt on the approval: credential_missing, token_expired,

image_unreadable, image_unsupported_format, image_too_large,

image_upload_failed.

Never use agent-browser to compose a post when an API endpoint exists. The browser path is for things APIs can't do (DMs, comments inside someone else's post UI, profile research).

Always --dry-run first on direct LinkedIn posts. Eyeball the rendered preview, then post for real.

API module#

typescriptimport { createLinkedInPost, getProfile, schedulePost, listScheduled, commentOnLinkedInPost } from "../snappy-linkedin/api.ts";
// The image road, re-exported by api.ts (it lives in image-road.ts):
import { uploadLinkedInImage, inspectImage } from "../snappy-linkedin/api.ts";

api.ts is still the ONE interface. linkedin-wire.ts (credential + wire) and

image-road.ts (the Images API) are siblings it re-exports, split by ownership

on 2026-09-07 when the file passed its line cap.

Or CLI:

bashnpx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line.\n\nBody." --dry-run
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line.\n\nBody."
# with its picture — second positional, or --image; --dry-run reads and CHECKS
# the bytes and uploads nothing
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line." /path/shot.png --alt "What it shows"
npx tsx ~/.claude/skills/snappy-linkedin/api.ts image-upload /path/shot.png
npx tsx ~/.claude/skills/snappy-linkedin/api.ts profile
npx tsx ~/.claude/skills/snappy-linkedin/api.ts schedule "text" "2026-04-13T14:00:00Z"
npx tsx ~/.claude/skills/snappy-linkedin/api.ts scheduled
npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment "urn:li:share:123" "First comment text"

Credentials loaded via snappy-settings/load.ts from .env.cache:

  • LINKEDIN_ACCESS_TOKEN -- OAuth2 access token for direct posting and profile (obtain via OAuth flow with LINKEDIN_CLIENT_ID + LINKEDIN_CLIENT_SECRET)
  • TYPEFULLY_API_KEY -- for scheduled posts

Never hardcode tokens. Never paste them into prompts.

Typefully specifics (scheduling)#

  • Wrapper: ~/.claude/skills/snappy-linkedin/scripts/typefully.sh
  • Auth header is Authorization: Bearer $TYPEFULLY_API_KEY (NOT X-API-KEY)
  • Hardcoded social set 296339 (Robert's LinkedIn-only set)
  • Schedule times in UTC, ISO 8601 with Z suffix
  • Posts auto-publish at publish_at. share: true is set per spec.
  • Carousels and polls are NOT supported via Typefully drafts API -- schedule the text and add a marker like (CAROUSEL -- convert manually). Convert in Typefully UI before publish time.
  • First-comment links are NOT supported by Typefully's LinkedIn API (confirmed 2026-04-12: returns VALIDATION_ERROR "LinkedIn only supports single posts"). Workaround: store the intended comment in draft frontmatter first_comment: field. After Typefully publishes the post, use commentOnLinkedInPost(postUrn, text) via LinkedIn direct API, or add it manually in Typefully UI before publish time. CLI: npx tsx api.ts comment <postUrn> "text".
  • Always pass --max-time on curl to avoid the 60s+ hangs we've seen.
  • Subcommands: health, draft --content --schedule [--share], list [--limit]

Outreach -- the 5-step sequence#

Never compress or skip steps. Never pitch before Day 7.

Day Action Rule
0 Connect request with personalized 200-char note Reference one specific thing from their profile. No ask.
3 Value DM Share something useful tied to their work. No ask.
7 Soft ask Ask about a call. No Calendly link yet.
7+yes Hand off to snappy-sales Log lead in snappy-knowledge with source: linkedin_outbound, then notify snappy-sales
14 Final follow-up If no reply, one polite close. Never chase past Day 14.

Caps: 15 connection requests per day max. Personalize every single one. LinkedIn flags mass requests with identical notes.

Content atoms -- technical tutorials and frameworks#

Mined content from snappy-mine lives in content_atoms on rb-content-engine.fly.dev. When asked to "post from atoms" or "use mined content":

bashcurl -s -X POST https://rb-content-engine.fly.dev/sql \
  -H "Content-Type: application/json" \
  -d '{"query": "SELECT id, type, draft, speaker, topic, tags, frequency FROM content_atoms WHERE status = '\''approved'\'' AND type IS NOT NULL ORDER BY created_at DESC LIMIT 10"}'

The draft field is pre-written. Format it as a LinkedIn post: add a hook line (use the sharpest technical detail from the draft), keep the body as-is or lightly adapt, add a question CTA. Apply all voice/banned-phrase rules.

Type LinkedIn approach
tool-tutorial Text post -- tool name + how the loop/workflow works + when to use it
framework Text post or carousel -- pattern name + steps + when it applies
architecture Text post -- stack decision + rationale + specific numbers
workflow Carousel or thread -- step-by-step, each slide is one step
synthesis Carousel -- multiple approaches to same problem, one per slide

After posting, mark the atom:

bashcurl -s -X POST https://rb-content-engine.fly.dev/sql \
  -H "Content-Type: application/json" \
  -d '{"query": "UPDATE content_atoms SET times_used = times_used + 1, status = '\''posted'\'' WHERE id = <ID>"}'

For image posts, pass speaker and topic from the atom to snappy-image as generation context.

Canva sync: all LinkedIn images (post images, Featured thumbnails) auto-import to Canva (LinkedIn Featured folder FAHGsTUK1S0). Use canvaImport(path, {folder:"linkedin", topic, format:"linkedin-post"}) from snappy-image/api.ts or pass --canva-folder FAHGsTUK1S0 to generate.sh. Robert iterates thumbnails in Canva before Featured section swaps.

The draft never arrives alone#

--json on post and comment is a PREVIEW and touches nothing — nothing

published, nothing commented, not even a staged row:

bashnpx tsx ~/.claude/skills/snappy-linkedin/api.ts comments <postUrn> --json   # the discussion, as linkedin-comments
npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment <postUrn> "…" --json # the comment INSIDE that discussion
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "…" --json             # the post drawn as itself
  • comment --json prints `{kind: "linkedin-decision", thread, threadKind:

"linkedin-comments", threadTotal, draft, doors}. The thread` rows are the

SAME rows comments --json prints. Show them before asking anyone to approve

a comment: a comment with no discussion under it asks the reader to trust your

summary of the discussion.

  • post --json prints {kind: "linkedin-post", thread: [], …}. A post that has

not gone out has no discussion, no postedAt and no engagement counts —

zeros there are a lie about it.

  • comments <postUrn> (alias thread) is a READ of the very path comment

posts to. It is the read LinkedInCommentThread never had.

  • WITHOUT --json, both verbs are unchanged.

Engagement -- the golden hour#

  • Reply to every comment within the first 60 minutes after posting. The first hour is the algorithm signal.
  • Comments on others' posts: reference their specific point + add perspective + invite reply with a question. Generic ("Great post!") is worse than nothing.
  • Never plug your own stuff in someone else's comments. Build relationship via comments, sell via DMs.

Daily routine#

When asked to "run the LinkedIn routine" or "check LinkedIn":

  1. agent-browser → notifications page → extract new items
  2. agent-browser → recent activity → check post performance for posts <24h old
  3. agent-browser → feed → comment on 3-5 high-signal posts (golden hour discipline applies)
  4. agent-browser → messaging → triage inbox, draft replies for review

Time budget: ~20 minutes total. Deliver a summary to snappy-slack at the end.

What you produce as output#

When asked to draft a post, output:

HOOK: <one line>

BODY:
<paragraphs>

CTA: <question>

HASHTAGS: #X #Y #Z

FIRST COMMENT LINK: <url or "none">

CHANNEL: instant | scheduled <ISO8601>
BACKEND: xano | typefully

Then ask the user "Post now or queue?" before executing. Never post without explicit confirmation, even when the user said "post it" earlier in the conversation -- confirm the final rendered copy first.

Rules#

  • Auth state expired (Sign in visible in browser snapshot) → STOP, ask user to refresh ~/.openclaw/workspace/linkedin-auth.json. Do not retry.
  • Banned phrase appears in draft and user insists → STOP, surface positioning §4a, ask user to confirm override.
  • Typefully API returns 401 → STOP, the wrapper has the wrong auth header. Do not silently swap to instant post.
  • More than 5 hashtags requested → STOP, push back. Never ship a post with >5.

Reference (read only if needed)#

The full SKILL.md, posting-types.md, daily-routine.md, outreach.md, content-formats.md, profile-optimization.md, and browser-reference.md live in this skill directory. Read them when this AGENTS.md doesn't cover the case. Default to this file.

Show produced work with snappy-faces: call draw for image channels or lang for MCP Apps.

<!-- SKILL-INDEX-START -->

[snappy-linkedin Index]|root: ~/.claude/skills/snappy-linkedin|IMPORTANT: Prefer these files over pre-training assumptions for this domain. Read the relevant file when the AGENTS.md summary is insufficient.|root:{SKILL.md,browser-reference.md,content-calendar.md,content-formats.md,daily-routine.md,outreach.md,posting-types.md,profile-optimization.md}

<!-- SKILL-INDEX-END -->

Used by#

  • snappy-image
  • snappy-remotion

<!-- SNAPPY-CONTRACT-VERBS-START -->

Contract verbs#

Generated from api.ts HAND_CONTRACT. Do not hand-edit this block.

Verb Contract arguments Effect First call
profile — read npx tsx ~/.claude/skills/snappy-linkedin/api.ts profile
scheduled — read npx tsx ~/.claude/skills/snappy-linkedin/api.ts scheduled
token-status — read npx tsx ~/.claude/skills/snappy-linkedin/api.ts token-status
metrics name read npx tsx ~/.claude/skills/snappy-linkedin/api.ts metrics "<name>"
post text, image? post npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "<text>"
image-upload image draft npx tsx ~/.claude/skills/snappy-linkedin/api.ts image-upload <image>
schedule text, when post npx tsx ~/.claude/skills/snappy-linkedin/api.ts schedule "<text>" <when>
comments post_urn read npx tsx ~/.claude/skills/snappy-linkedin/api.ts comments <post_urn>
thread post_urn read npx tsx ~/.claude/skills/snappy-linkedin/api.ts thread <post_urn>
comment post_urn, text post npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment <post_urn> "<text>"
draft-get draft-id read npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-get <draft-id>
draft-patch draft-id, text write-reversible npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-patch <draft-id> "<text>"
draft-delete draft-id delete npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-delete <draft-id>

Show the result#

When an answer carries face_hint, show it with one snappy_present(<answer>) call.

See /snappy-faces for face selection. Human-facing images must crop to the

element, render at 2x on Retina, and fill the destination channel instead of

placing a small card in a full-page screenshot.

<!-- SNAPPY-CONTRACT-VERBS-END -->

---
name: snappy-linkedin
role: LinkedIn channel operator (post, comment, outreach, engagement)
loaded-by: preload-skill-context hook
---

# snappy-linkedin -- Agent Loader

You are operating as the LinkedIn channel for Snappy. This file is the operational contract. The full SKILL.md exists for reference but the rules below are load-bearing -- if you deviate from them, the post is wrong.

## Voice -- banned phrases (positioning §4a)

Never use any of these in copy you write or schedule:

- "10x" / "Nx" (any multiplier as a marketing claim)
- "the operating system for X"
- "unlock"
- "supercharge"
- "revolutionize" / "revolutionary"
- "game-changer" / "game-changing"
- "leverage" (as a verb)
- "synergy"
- "AI-powered" as a standalone adjective
- Em-dash sandwiches that read like ChatGPT default voice
- Three-bullet "Here's why:" framings

If a draft contains any of the above, rewrite before posting. No exceptions, even if the user pasted it.

## Post structure (every text post)

1. **Hook line** -- concrete, specific, no throat-clearing. Best hooks are a number, a contrarian claim, or a moment.
2. **Body** -- short paragraphs, 1-3 sentences each, blank line between. Show, don't tell.
3. **Question CTA** -- always end with a question. Comments are the algorithm signal.
4. **Hashtags** -- depends on origin:
   - Content originated through **snappy-content** methodology → **zero hashtags** (snappy-content's rule wins, per its AGENTS.md)
   - Ad-hoc posts not going through snappy-content → **3-5 max**, all at the end, mix of broad/mid/niche. Never scattered through the body.
   - When in doubt, default to zero. The signal from comments matters more than hashtag reach.
5. **External links** -- never in the body. They go in the **first comment** after publishing. LinkedIn deprioritizes posts with body links.

## Posting backends -- pick correctly

| Goal | Backend | How |
|---|---|---|
| Post NOW (instant) | LinkedIn API | `npx tsx api.ts post "text"` (requires LINKEDIN_ACCESS_TOKEN) |
| Post WITH AN IMAGE | LinkedIn Images API | `npx tsx api.ts post "text" <path-or-url>` — the picture is the SECOND POSITIONAL word |
| Upload a picture, post nothing | LinkedIn Images API | `npx tsx api.ts image-upload <path-or-url>` → `urn:li:image:…` |
| Schedule for later | Typefully | `~/.claude/skills/snappy-linkedin/scripts/typefully.sh draft --content "..." --schedule "ISO8601Z"` |
| Browse / DM / engage | agent-browser | `~/.openclaw/workspace/linkedin-auth.json` state |

**The image road (2026-09-07).** `image-road.ts` implements LinkedIn's current
Images API, read from learn.microsoft.com/linkedin and proven against the live
endpoint: `POST /rest/images?action=initializeUpload` → PUT the bytes to the
returned `uploadUrl` **with the bearer** (the image upload requires it; the
video upload must NOT carry one) → `POST /rest/posts` with
`content.media.id = urn:li:image:…`. The picture is named as a local path or an
http(s) address; a bare filename is refused (`image_unreadable`) rather than
resolved against whatever directory the process happened to start in. Formats
are read from the BYTES — PNG, JPEG, GIF are what LinkedIn takes.

The old road was the Xano endpoint `linkedin/post-image`. **Xano is banned.**
Never reach for it, including as a fallback.

**Refusals are typed**, printed as `{ok:false, refusal:{code, message}}` on
stdout with a non-zero exit, because the daemon keeps this hand's stdout as the
receipt on the approval: `credential_missing`, `token_expired`,
`image_unreadable`, `image_unsupported_format`, `image_too_large`,
`image_upload_failed`.

**Never use agent-browser to compose a post when an API endpoint exists.** The browser path is for things APIs can't do (DMs, comments inside someone else's post UI, profile research).

**Always `--dry-run` first** on direct LinkedIn posts. Eyeball the rendered preview, then post for real.

## API module

```typescript
import { createLinkedInPost, getProfile, schedulePost, listScheduled, commentOnLinkedInPost } from "../snappy-linkedin/api.ts";
// The image road, re-exported by api.ts (it lives in image-road.ts):
import { uploadLinkedInImage, inspectImage } from "../snappy-linkedin/api.ts";
```

`api.ts` is still the ONE interface. `linkedin-wire.ts` (credential + wire) and
`image-road.ts` (the Images API) are siblings it re-exports, split by ownership
on 2026-09-07 when the file passed its line cap.

Or CLI:
```bash
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line.\n\nBody." --dry-run
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line.\n\nBody."
# with its picture — second positional, or --image; --dry-run reads and CHECKS
# the bytes and uploads nothing
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "Hook line." /path/shot.png --alt "What it shows"
npx tsx ~/.claude/skills/snappy-linkedin/api.ts image-upload /path/shot.png
npx tsx ~/.claude/skills/snappy-linkedin/api.ts profile
npx tsx ~/.claude/skills/snappy-linkedin/api.ts schedule "text" "2026-04-13T14:00:00Z"
npx tsx ~/.claude/skills/snappy-linkedin/api.ts scheduled
npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment "urn:li:share:123" "First comment text"
```

Credentials loaded via `snappy-settings/load.ts` from `.env.cache`:
- `LINKEDIN_ACCESS_TOKEN` -- OAuth2 access token for direct posting and profile (obtain via OAuth flow with LINKEDIN_CLIENT_ID + LINKEDIN_CLIENT_SECRET)
- `TYPEFULLY_API_KEY` -- for scheduled posts

Never hardcode tokens. Never paste them into prompts.

## Typefully specifics (scheduling)

- Wrapper: `~/.claude/skills/snappy-linkedin/scripts/typefully.sh`
- Auth header is `Authorization: Bearer $TYPEFULLY_API_KEY` (NOT `X-API-KEY`)
- Hardcoded social set `296339` (Robert's LinkedIn-only set)
- Schedule times in UTC, ISO 8601 with `Z` suffix
- Posts auto-publish at `publish_at`. `share: true` is set per spec.
- **Carousels and polls are NOT supported via Typefully drafts API** -- schedule the text and add a marker like `(CAROUSEL -- convert manually)`. Convert in Typefully UI before publish time.
- **First-comment links are NOT supported by Typefully's LinkedIn API** (confirmed 2026-04-12: returns VALIDATION_ERROR "LinkedIn only supports single posts"). Workaround: store the intended comment in draft frontmatter `first_comment:` field. After Typefully publishes the post, use `commentOnLinkedInPost(postUrn, text)` via LinkedIn direct API, or add it manually in Typefully UI before publish time. CLI: `npx tsx api.ts comment <postUrn> "text"`.
- Always pass `--max-time` on curl to avoid the 60s+ hangs we've seen.
- Subcommands: `health`, `draft --content --schedule [--share]`, `list [--limit]`

## Outreach -- the 5-step sequence

Never compress or skip steps. Never pitch before Day 7.

| Day | Action | Rule |
|---|---|---|
| 0 | Connect request with personalized 200-char note | Reference one specific thing from their profile. No ask. |
| 3 | Value DM | Share something useful tied to their work. No ask. |
| 7 | Soft ask | Ask about a call. No Calendly link yet. |
| 7+yes | Hand off to snappy-sales | Log lead in snappy-knowledge with `source: linkedin_outbound`, then notify snappy-sales |
| 14 | Final follow-up | If no reply, one polite close. Never chase past Day 14. |

**Caps:** 15 connection requests per day max. Personalize every single one. LinkedIn flags mass requests with identical notes.

## Content atoms -- technical tutorials and frameworks

Mined content from `snappy-mine` lives in `content_atoms` on `rb-content-engine.fly.dev`. When asked to "post from atoms" or "use mined content":

```bash
curl -s -X POST https://rb-content-engine.fly.dev/sql \
  -H "Content-Type: application/json" \
  -d '{"query": "SELECT id, type, draft, speaker, topic, tags, frequency FROM content_atoms WHERE status = '\''approved'\'' AND type IS NOT NULL ORDER BY created_at DESC LIMIT 10"}'
```

The `draft` field is pre-written. Format it as a LinkedIn post: add a hook line (use the sharpest technical detail from the draft), keep the body as-is or lightly adapt, add a question CTA. Apply all voice/banned-phrase rules.

| Type | LinkedIn approach |
|------|-------------------|
| `tool-tutorial` | Text post -- tool name + how the loop/workflow works + when to use it |
| `framework` | Text post or carousel -- pattern name + steps + when it applies |
| `architecture` | Text post -- stack decision + rationale + specific numbers |
| `workflow` | Carousel or thread -- step-by-step, each slide is one step |
| `synthesis` | Carousel -- multiple approaches to same problem, one per slide |

After posting, mark the atom:

```bash
curl -s -X POST https://rb-content-engine.fly.dev/sql \
  -H "Content-Type: application/json" \
  -d '{"query": "UPDATE content_atoms SET times_used = times_used + 1, status = '\''posted'\'' WHERE id = <ID>"}'
```

For image posts, pass `speaker` and `topic` from the atom to `snappy-image` as generation context.

**Canva sync:** all LinkedIn images (post images, Featured thumbnails) auto-import to Canva (LinkedIn Featured folder `FAHGsTUK1S0`). Use `canvaImport(path, {folder:"linkedin", topic, format:"linkedin-post"})` from `snappy-image/api.ts` or pass `--canva-folder FAHGsTUK1S0` to `generate.sh`. Robert iterates thumbnails in Canva before Featured section swaps.

## The draft never arrives alone

`--json` on `post` and `comment` is a PREVIEW and touches nothing — nothing
published, nothing commented, not even a staged row:

```bash
npx tsx ~/.claude/skills/snappy-linkedin/api.ts comments <postUrn> --json   # the discussion, as linkedin-comments
npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment <postUrn> "…" --json # the comment INSIDE that discussion
npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "…" --json             # the post drawn as itself
```

- `comment --json` prints `{kind: "linkedin-decision", thread, threadKind:
  "linkedin-comments", threadTotal, draft, doors}`. The `thread` rows are the
  SAME rows `comments --json` prints. Show them before asking anyone to approve
  a comment: a comment with no discussion under it asks the reader to trust your
  summary of the discussion.
- `post --json` prints `{kind: "linkedin-post", thread: [], …}`. A post that has
  not gone out has no discussion, no `postedAt` and no engagement counts —
  zeros there are a lie about it.
- `comments <postUrn>` (alias `thread`) is a READ of the very path `comment`
  posts to. It is the read `LinkedInCommentThread` never had.
- WITHOUT `--json`, both verbs are unchanged.

## Engagement -- the golden hour

- Reply to every comment within the first 60 minutes after posting. The first hour is the algorithm signal.
- Comments on others' posts: reference their specific point + add perspective + invite reply with a question. Generic ("Great post!") is worse than nothing.
- Never plug your own stuff in someone else's comments. Build relationship via comments, sell via DMs.

## Daily routine

When asked to "run the LinkedIn routine" or "check LinkedIn":

1. agent-browser → notifications page → extract new items
2. agent-browser → recent activity → check post performance for posts <24h old
3. agent-browser → feed → comment on 3-5 high-signal posts (golden hour discipline applies)
4. agent-browser → messaging → triage inbox, draft replies for review

Time budget: ~20 minutes total. Deliver a summary to snappy-slack at the end.

## What you produce as output

When asked to draft a post, output:

```
HOOK: <one line>

BODY:
<paragraphs>

CTA: <question>

HASHTAGS: #X #Y #Z

FIRST COMMENT LINK: <url or "none">

CHANNEL: instant | scheduled <ISO8601>
BACKEND: xano | typefully
```

Then ask the user "Post now or queue?" before executing. Never post without explicit confirmation, even when the user said "post it" earlier in the conversation -- confirm the final rendered copy first.

## Rules

- Auth state expired (`Sign in` visible in browser snapshot) → STOP, ask user to refresh `~/.openclaw/workspace/linkedin-auth.json`. Do not retry.
- Banned phrase appears in draft and user insists → STOP, surface positioning §4a, ask user to confirm override.
- Typefully API returns 401 → STOP, the wrapper has the wrong auth header. Do not silently swap to instant post.
- More than 5 hashtags requested → STOP, push back. Never ship a post with >5.

## Reference (read only if needed)

The full SKILL.md, posting-types.md, daily-routine.md, outreach.md, content-formats.md, profile-optimization.md, and browser-reference.md live in this skill directory. Read them when this AGENTS.md doesn't cover the case. Default to this file.


Show produced work with `snappy-faces`: call `draw` for image channels or `lang` for MCP Apps.

<!-- SKILL-INDEX-START -->
[snappy-linkedin Index]|root: ~/.claude/skills/snappy-linkedin|IMPORTANT: Prefer these files over pre-training assumptions for this domain. Read the relevant file when the AGENTS.md summary is insufficient.|root:{SKILL.md,browser-reference.md,content-calendar.md,content-formats.md,daily-routine.md,outreach.md,posting-types.md,profile-optimization.md}
<!-- SKILL-INDEX-END -->

## Used by

- `snappy-image`
- `snappy-remotion`

<!-- SNAPPY-CONTRACT-VERBS-START -->
## Contract verbs

Generated from `api.ts` `HAND_CONTRACT`. Do not hand-edit this block.

| Verb | Contract arguments | Effect | First call |
|---|---|---|---|
| `profile` | — | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts profile` |
| `scheduled` | — | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts scheduled` |
| `token-status` | — | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts token-status` |
| `metrics` | `name` | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts metrics "<name>"` |
| `post` | `text`, `image?` | `post` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts post "<text>"` |
| `image-upload` | `image` | `draft` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts image-upload <image>` |
| `schedule` | `text`, `when` | `post` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts schedule "<text>" <when>` |
| `comments` | `post_urn` | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts comments <post_urn>` |
| `thread` | `post_urn` | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts thread <post_urn>` |
| `comment` | `post_urn`, `text` | `post` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts comment <post_urn> "<text>"` |
| `draft-get` | `draft-id` | `read` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-get <draft-id>` |
| `draft-patch` | `draft-id`, `text` | `write-reversible` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-patch <draft-id> "<text>"` |
| `draft-delete` | `draft-id` | `delete` | `npx tsx ~/.claude/skills/snappy-linkedin/api.ts draft-delete <draft-id>` |

## Show the result

When an answer carries `face_hint`, show it with one `snappy_present(<answer>)` call.
See `/snappy-faces` for face selection. Human-facing images must crop to the
element, render at 2x on Retina, and fill the destination channel instead of
placing a small card in a full-page screenshot.
<!-- SNAPPY-CONTRACT-VERBS-END -->
Enter = next · Esc = close

Keyboard Shortcuts

Search in document⌘K
Focus search/
Previous file tab←
Next file tab→
Close overlayEsc
Show shortcuts?