linkedin-likes-only

Cron job (weekdays 9 AM). LIKES ONLY -- no comments, no replies, no posting.

Hard rules

• NEVER write comments as Robert. This job was explicitly scoped to likes only.
• NEVER post anything. No new posts, no replies, no DMs.
• NEVER use agent-browser to fill any text field or click any "Post"/"Reply"/"Send" button.
• The ONLY interaction allowed is clicking the Like button on posts.

Steps

1. Set session isolation:

   export AGENT_BROWSER_SESSION="linkedin-likes-$$-$(date +%s)"

1. Open LinkedIn feed with auth state:

   agent-browser --state ~/.agent-browser/sessions/linkedin-auth.json open "https://www.linkedin.com/feed"

1. Wait for page load:

   agent-browser wait 3000

1. Check auth -- if "Sign in" is visible in a snapshot, STOP immediately and send a Telegram alert that auth is expired. Do not proceed.

1. Take a snapshot and extract the first 15-20 posts from the feed:

   agent-browser snapshot -i

1. Pick 5-10 posts to like. Priority:

• Posts from people in Robert's network (clients, prospects, collaborators)
• Posts about AI agents, Claude Code, MCP, agentic building, developer tools
• Posts from people who recently engaged with Robert's content
• Skip: ads, job postings, reshares with no original commentary, viral meme content

1. For each selected post, click the Like button using the @ref from the snapshot:

   agent-browser click @eN   # where @eN is the Like button ref
   agent-browser wait 500

Re-snapshot between likes if refs go stale.

1. Keep a log of each liked post: author name, topic/first line, and the ref used.

1. Save auth state back before closing:

   agent-browser state save ~/.agent-browser/sessions/linkedin-auth.json

1. Close the session:

    agent-browser close

Report

After completing likes, send a Telegram summary:

npx tsx state/lib/telegram.ts send "*LinkedIn Likes*

Liked $(COUNT) posts:
$(LIST_OF_LIKED_POSTS_WITH_AUTHOR_AND_TOPIC)

Auth status: OK
Session: $(date '+%Y-%m-%d %H:%M')"

Auth failure

If auth is expired at any point, send this Telegram alert and exit:

npx tsx state/lib/telegram.ts send "*FAILED* LinkedIn Likes -- auth expired. Re-auth needed."

Eval

Score 1.0 if >= 5 posts liked and Telegram summary sent. Score 0.5 if < 5 posts liked. Score 0.0 if auth failed or no posts liked.

Rubric

criteria:
  - name: no_forbidden_actions
    kind: deterministic
    check: "Verify that no 'agent-browser fill', 'agent-browser click' on 'Post'/'Reply'/'Send' buttons, or any other write operations occurred in the agent-browser logs."
  - name: posts_liked_count
    kind: deterministic
    check: "Check the Telegram message content for 'Liked $(COUNT) posts' and confirm that 'COUNT' is >= 5."
  - name: linkedin_auth_status
    kind: deterministic
    check: "Verify the Telegram message contains 'Auth status: OK' and no 'FAILED* LinkedIn Likes -- auth expired' message was sent."
  - name: relevant_posts_selection
    kind: judge
    check: "Review the 'LIST_OF_LIKED_POSTS_WITH_AUTHOR_AND_TOPIC' in the Telegram summary to ensure selected posts align with priority criteria (e.g., Robert's network, AI topics, not ads/job postings)."

linkedin-likes-only - loader

Per-turn rules for the linkedin-likes-only skill. Full reference: state/skills/linkedin-likes-only/SKILL.md. Do not skip these.

Critical Rules

• NEVER write comments as Robert - this job was explicitly scoped to likes only
• NEVER post anything - no new posts, no replies, no DMs
• NEVER use agent-browser to fill any text field or click Post/Reply/Send
• The ONLY allowed interaction is clicking the Like button on selected posts
• ALWAYS isolate the agent-browser session: export AGENT_BROWSER_SESSION="linkedin-likes-$$-$(date +%s)"
• ALWAYS pass --state ~/.agent-browser/sessions/linkedin-auth.json on agent-browser open, NOT via state load after launch
• If "Sign in" appears in a snapshot → STOP, send Telegram alert about expired auth, do not proceed
• ALWAYS save auth state back before closing the session

Commands

| ui dashboard | state/skills/linkedin-likes-only/resources/ui.openui | |invoke: skill is prose - follow steps 1-10 in state/skills/linkedin-likes-only/SKILL.md |open feed: agent-browser --state ~/.agent-browser/sessions/linkedin-auth.json open "https://www.linkedin.com/feed" |like a post: agent-browser click @eN (where @eN is the Like button ref from the snapshot) |telegram report: npx tsx state/lib/telegram.ts send "*LinkedIn Likes* ..." |eval log: state/log/evals.ndjson (skill: "linkedin-likes-only")

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/linkedin-likes-only/resources/ui.openui. Read it before rendering or editing this skill's generated component surface.
• Treat this resource as a first-class artifact of the skill, not a generic chat response. Improve it when the skill's user-facing output needs to become richer.
• System resources compose OpenUI primitives and inherit SnappyChat tokens. Use ui_contract: branded in SKILL.md only for deliberate platform or client visuals.

Known Pitfalls

• Refs go stale between actions - re-snapshot between likes
• Pick 5-10 posts max from network/AI-agent topics; skip ads, job postings, no-commentary reshares, viral memes
• Score: 1.0 if ≥5 liked + Telegram summary sent; 0.5 if <5; 0.0 if auth failed or zero likes

Self-Test

An agent reading this should correctly:

1. [ ] Refuse to comment on a post even if the user asks for "engagement"?
2. [ ] Set AGENT_BROWSER_SESSION before opening LinkedIn?
3. [ ] Halt and send a Telegram alert when "Sign in" appears in a snapshot?

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] linkedin-likes-only: <what was missing>" >> state/log/loader-feedback.log

<!-- kernel-ok: historical attribution for the PID-loader pattern. -->

Self-correcting loader (PID feedback)

This is the rule that makes loaders learn. Ported from snappy-kernel, where it demonstrably worked: the snappy-dom-cartographer skill mapped Skool admin UI once, every later ship-agent inherited the full map for free.

Before you finish, do two things:

1. Fix gaps (P - proportional). MANDATORY. If this loader didn't cover your case - if you had to read another file, run an undocumented command, work around a wrong selector, or discover a quirk - you MUST attempt an Edit to this AGENTS.md before you log.

• Default action: edit this file. Surgical. Add the missing line, table

row, or code example where it belongs. No restructuring.

• Only LOGGED is allowed when: the fix needs >10 lines, spans multiple

files, or requires a structural rewrite. In that case the state/regen/drain.sh queue picks it up asynchronously.

• "I didn't have time" / "it's minor" / "the next agent will figure it out"

are NOT valid reasons. Minor gaps compound - that's the whole point of this loop.

• Concrete example of a valid P-fix: discover agent-browser screenshot

takes path positionally not as --output <path>. One line in a table. Edit it in. Do not just log it.

• The goal: the next agent never has to leave the loader.

2. Log the result.

echo "[$(date -u +%FT%TZ)] linkedin-likes-only: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/loader-feedback.log

• FIXED = you patched this loader inline (P-fix).
• LOGGED = too large for inline; the PostToolUse enqueue + Stop-hook drain

will rewrite the loader from scratch on next session-end.

Do not skip this. Every agent run must leave the system better than it found it. The loader is the setpoint; you are the sensor; the gap is the error signal; closing the gap is the correction.