Notre avis

Automatise les commentaires sur LinkedIn avec un ton naturel et des contraintes anti-détection (navigation headfull, délais aléatoires, interdiction des phrases interdites).

Points forts

Respecte des contraintes strictes de type humain (vitesse de frappe, délais aléatoires) pour éviter la détection.
Impose une limite quotidienne (20 commentaires/UTC) et une vérification de doublons inter-sessions.
Utilise un profil de session persistant pour éviter de se reconnecter à chaque exécution.

Limites

Nécessite une première connexion manuelle sur LinkedIn.
Dépend de la non-détection des techniques de stealth (risque d'obsolescence si LinkedIn met à jour ses protections).
Ne gère pas les commentaires sur des posts très récents ou les réponses à des commentaires existants.

Quand l'utiliser

Pour poster de manière régulière et discrète des commentaires pertinents sur LinkedIn sans intervention manuelle, par exemple pour une stratégie de marketing ou de networking.

Quand l'éviter

Si vous avez besoin de commenter plus de 20 fois par jour ou de réagir à des conversations en temps réel, car les limites et délais rendent le processus trop lent.

Exemples

Comment on latest industry post

Run the LinkedIn auto-commenter to leave a thoughtful comment on the latest post from my feed about AI in healthcare, following all stealth and quality rules.

Schedule 5 comments today

Use the LinkedIn auto-commenter to post 5 comments today, spaced at least 15 seconds apart, on posts about remote work trends. Make sure each comment references a specific detail from the post and adds value.

Check daily cap and comment if under limit

Check if the LinkedIn auto-commenter has remaining daily comments (under 20) and if so, comment on the post at https://www.linkedin.com/posts/example-post with a natural, conversational response that does not start with a compliment.

LinkedIn Auto-Commenter — Project Standards

Comment Quality

Tone

Conversational and direct — write like a knowledgeable peer, not a marketing department
2–4 sentences per comment
Use contractions naturally (it's, I'm, that's, we've, etc.)
Reference specific details from the post — generic comments are not acceptable
Add value: a question, a relevant experience, a constructive perspective

Prohibited Content

Never start a comment with a compliment about the post ("Great post!", "So insightful!")
Never include any phrase from src/ai/banned_phrases.py
Never reveal AI, automated, or generated origins
No corporate buzzwords: leverage, synergy, paradigm shift, game-changer, etc.

Operational Limits

Daily Cap

Hard limit: 20 comments per UTC day
Enforced by RateLimiter.assert_can_post() before every comment attempt
DailyLimitExceededError is raised — do not bypass or catch and ignore

Timing

Minimum delay between comments: 15 seconds
Maximum delay between comments: 45 seconds
Typing speed: 55–80 WPM with +/-30% per-character jitter

Architecture Constraints

Stealth Requirements

Browser must run headful (headless=False) — LinkedIn detects headless via GPU/font fingerprinting
stealth_async(context) must be called AFTER launch_persistent_context() returns
Use STEALTH_BROWSER_ARGS from browser_factory.py — do not modify without testing

Data Immutability

All models use frozen=True (@dataclass(frozen=True) or Pydantic frozen=True)
Never mutate model instances — create new ones instead

Duplicate Prevention

ActivityLog.was_commented(post_url) is checked before adding any post to the scrape results
Do not comment on the same post URL twice across sessions

Persistent Session

Browser profile stored at data/browser_profile/ (gitignored)
First run: log into LinkedIn manually in the opened browser
Subsequent runs: session is restored automatically — no re-login needed

File Organization

Max 800 lines per file, target 200–400 lines
One concern per module — no mega-files
All error handling must be explicit — never swallow exceptions silently

LinkedIn Auto-Commenteur

Recommandé pour