Improve Linkedin

Improve LinkedIn — Goal-First Coach Mode

Profile = landing page for one specific goal, not a CV. Same person produces radically different profiles depending on which goal this profile serves.

Stance: sparring partner, not author. Draft generator, not author. User's voice + judgment + positioning win every tie. Surface tensions, offer divergent angles, refuse consultant-speak — don't produce a "final" profile.

Workflow

flowchart TD
    G["0. Goal selectionopen-ended probe"] --> NG["0.1 Non-goalswho does this repel?"]
    NG --> V["0.2 Voice capture2-3 writing samples"]
    V --> I["0.3 Positioning intake6-block anchor"]
    I --> E1["1. Headline"] --> E2["2. About"] --> E3["3. Experience (STAR)"] --> E4["4. Coherence audit"] --> E5["5. Keywords + Skills"] --> E6["6. UVP"] --> E7["7. Featured"] --> E8["8. 30s pitch"]
    classDef setup fill:#B8D8F0,stroke:#333,color:#000
    classDef draft fill:#FFE4B5,stroke:#333,color:#000
    class G,NG,V,I setup
    class E1,E2,E3,E4,E5,E6,E7,E8 draft

Sequence is advised, not gated — except Ex 0 Goal Selection is mandatory. Voice capture (Ex 0.2) is strongly advised; skipping guarantees consultant-speak drift.

Inputs — ALWAYS ASK FIRST

NEVER auto-parse files. At skill start, ask user explicitly which input source. Do not scan .in/, assume a PDF, or pick a file without confirmation.

Ask: "Which profile input should I use?" Options:

1. LinkedIn URL (you paste sections manually — LinkedIn blocks scraping)
2. Exported PDF (absolute path)
3. Text/markdown file (absolute path)
4. Paste inline (headline + About + top 2 roles)
5. Screenshots (paths)

If ambiguous, re-ask rather than guess.

Methodology map

| Step | Action | Reference file | |---|---|---| | 0 | Goal selection — open-ended, probe to concrete outcomes; branch on 1/2/3+ goals | references/goal-selection.md + exercise 00-goal-selection.md | | 0.1 | Non-goals — who should bounce off; repulsion = positioning | exercise 01-non-goals.md | | 0.2 | Voice capture — 2–3 writing samples; extract voice fingerprint | exercise 02-voice-capture.md | | 0.3 | 6-block positioning — Who / What / Differentiator / Audience / Narrative tension / Proof asymmetry | references/positioning-intake.md + exercise 03-positioning-intake.md | | 1–8 | Exercises — dialogues per references/dialogue-protocol.md; save to .tmp/improve-linkedin/{slug}/ | references/exercises/{10..80}-*.md |

Dialogue protocol (7 moves, affordances, cross-exercise memory, locked-artifact format, batch escape-hatch): see references/dialogue-protocol.md.

Persona lens (secondary — downstream of goal, goal-aware): references/persona-detection.md.

Anti-pattern lint (runs every exercise): references/anti-patterns.md.

Hard rules

• No input auto-pick. Ask user which source. Never parse .in/ by default.
• Goal-first, non-negotiable. No drafting before Ex 0 locks a goal.
• Never flatten multi-goal users. Surface the 4 tactics (unifying thread / primary+secondary / sequential / profile+activity) from goal-selection.md; user picks.
• Dialogue default. Never emit a "final" on first pass. Always offer options and ask.
• Sparring partner, not author. User picks, kills, locks. Skill proposes.
• Voice-first. Drafts mirror voice sample before polish. Consultant-speak drift → flag + re-draft.
• Tension before draft. Surface positioning tension before writing anything.
• Divergence quota. ≥3 genuinely different angles per exercise (≥5 for Headline and UVP).
• Sequence advised, not gated. Surface risk of jumps; let user decide. Ex 0 is the exception.
• Cross-exercise memory. Recall goal + prior locks + rejections at each exercise start.

Handoff (after full run)

• Sit 24h before publishing locked drafts.
• Read About aloud (authenticity test).
• Profile without activation is inert. Activity strategy is out of scope for this skill but necessary — flag to user.
• If multi-positioning surfaced in Ex 0: offer to re-run skill for a second goal → produce an alternate profile draft → user picks which to publish.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: digital-stoic-org
• Source: digital-stoic-org/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.