skool-partner-cta

Weekly Skool community post calling on members to complete the four Anthropic Skilljar courses. Anthropic accepted Snappy's Partner Network application; activation requires ten individuals completing the full Anthropic Academy learning path. This skill drafts the post + image, parks both in state/log/skool-partner-cta/, then waits for Robert to trigger the actual upload. No auto-posting.

The cadence (weekly) is set to ride alongside the existing Wednesday 1 PM ET Claude Code Meetup so completion proof and the meetup feed each other. The skill never modifies its own cron - it just produces a draft Robert can approve.

Steps

1. Scope - read recent Skool activity

• Read state/log/skool-partner-cta/ for the last 4 drafts (date stem)

and skip if today's date already has a draft.

• Read state/lib/skool-fetcher.ts recent posts (last 14 days) so the

draft can lightly reference prior completions if any have landed.

2. Render the image

Invoke the existing 6-layer pipeline. Do NOT freehand a prompt - use the validated illustration shape (cream paper, locked layers, black ink linework, four labeled squares for the four courses).

bash state/bin/image/generate-iterate.sh \
  --illustration "<6-layer fragment — see today's draft frontmatter>" \
  --text "Help us activate the Partner Network" \
  --slug "partner-cta-$(date +%Y-%m-%d)" \
  --out-dir "/tmp/snappy-image-partner-cta" \
  --format "linkedin-post" \
  --max-auto-turns 1

Read /tmp/snappy-image-partner-cta/<slug>/state.json. The last frame's verdict field is the gate - pass means write the image URL into the draft frontmatter; fail after max turns means log a friction and leave the image field as pending.

3. Compose the post body

Draft into state/log/skool-partner-cta/<YYYY-MM-DD>.md. Hard rules:

• Under 150 words.
• Names all four courses verbatim with their Skilljar URLs.
• Explains the concrete benefit of activating the Partner Network

(Partner Portal access, certification enrollment, full program path).

• No scarcity hooks ("idle hours", "leftover budget", "only X spots").
• No fake certificates / gamification language ("claim your certificate",

"level up", "earn your badge"). Value is understanding the SHAPE of the tech.

• No personal-story openers ("I was…", "Last week I…").
• Ends with a genuine question, not engagement-bait ("Agree?",

"Thoughts?" both fail the voice gate).

• Image-prompt frontmatter is §4a-exempt - em-dashes there are fine

because that text goes to Gemini, not the reader.

4. Gate the body through voice

import { checkTone } from "../lib/voice.ts";
const violations = checkTone(body);
if (violations.length) { /* loop until clean, never bypass */ }

A failing draft does NOT ship. It is rewritten and re-checked.

5. Wait for Robert to trigger send

The skill stops here. The draft sits in state/log/skool-partner-cta/ until Robert reviews and approves. A separate skill (or manual paste) handles the actual post via state/lib/community.ts createSkoolPost() or the Skool web session. Serialize the upload - no parallel browser writes against one Skool session (memory: no_parallel_browser_writes).

6. Log + eval

append("chain", { run_id, skill: "skool-partner-cta", action: "draft" });
score("skool-partner-cta", run_id, {
  score: imageVerdict === "pass" && voiceClean ? 1.0 : voiceClean ? 0.5 : 0.0,
  primary_issue: voiceClean ? null : "voice gate violation",
  image_status: imageVerdict,
  draft_path: `state/log/skool-partner-cta/${dateISO}.md`,
});

Eval

Actor: the dispatch model that drafts the body + the generate-iterate.sh Nano Banana 2 image draft. Auditor: voice.checkTone() for the body text + Gemini 3.1 Pro vision critic for the image (different model from the actor - clean actor/auditor split).

The skill is eval: auto because both auditors are deterministic and independent of the actor.

Gotchas

• The four Skilljar URLs are exact and must be reproduced verbatim. A

shortened or paraphrased URL fails the post's purpose. (See ~/.claude/projects/.../memory/project_claude_partner_program.md.)

• The Skool sidebar is flat - there is no classroom category for

"Partner Path". Don't model the post around a category that doesn't exist (memory: project_skool_no_categories).

• The course lives in Skool, not Slack. There is no separate Slack

group (memory: project_course_skool_not_slack). Any draft that points members to a Slack channel for the course is wrong on its face.

• The image-prompt fragment goes to Gemini, NEVER to a Skool reader.

Em-dashes there are fine. The post body is reader-facing and must pass checkTone() (memory: feedback_image_prompts_not_4a).

• Round-tripped drafts can show , (space-comma-two-spaces) where

em-dashes were stripped. Scan the literal string before posting (memory: feedback_skool_emdash_backfill_artifact).

Graduation

Stay prose until two consecutive runs draft cleanly without edits to the skill page itself. Then extract the body composer into state/bin/skool-partner-cta/run.ts and mark graduation: graduated. The image render already lives in graduated infra (state/bin/image/generate-iterate.sh); only the body composer needs to migrate.

Rubric

criteria:
  - name: draft_file_created
    kind: deterministic
    check: "A new Markdown file named 'YYYY-MM-DD.md' exists in 'state/log/skool-partner-cta/' reflecting the skill's execution date."
  - name: image_generated_and_logged
    kind: deterministic
    check: "The 'state.json' file for the generated image in '/tmp/snappy-image-partner-cta/<slug>/' shows a 'pass' verdict and its URL is present in the draft's frontmatter."
  - name: voice_gate_passed
    kind: deterministic
    check: "The log output or skill report explicitly states that 'checkTone(body)' reported no violations for the final draft."
  - name: post_body_content_quality
    kind: judge
    check: "The post body is under 150 words, accurately names all four Skilljar courses with correct URLs, explains the partner network benefits, avoids scarcity or gamification language, and ends with a genuine question, as detailed in the 'Compose the post body' section."

skool-partner-cta - loader

Per-turn rules for the skool-partner-cta skill. Full reference: state/skills/skool-partner-cta/SKILL.md. Do not skip these.

Critical Rules

• NEVER auto-post to Skool. The skill drafts to state/log/skool-partner-cta/<date>.md and STOPS. Robert triggers the upload.
• ALWAYS render the image via state/bin/image/generate-iterate.sh (the validated 6-layer pipeline). NEVER freehand a Gemini prompt or substitute a different model. (memory: feedback_image_quality_standard, feedback_image_system_validated)
• NEVER fake certificates / gamification language. No "claim your certificate", "earn your badge", "level up". Value = understanding the SHAPE of the tech. (memory: feedback_no_fake_certificates)
• The course lives in Skool, NOT Slack. There is no separate Slack group. Any draft pointing members to a Slack channel is wrong. (memory: project_course_skool_not_slack)
• ALL FOUR Skilljar URLs must appear verbatim - no paraphrase, no shortener:

1. https://anthropic.skilljar.com/introduction-to-agent-skills
2. https://anthropic.skilljar.com/claude-with-the-anthropic-api
3. https://anthropic.skilljar.com/introduction-to-model-context-protocol
4. https://anthropic.skilljar.com/claude-code-in-action

• ALWAYS run the body through voice.checkTone() before saving. Em-dashes, hype words, "Agree?" / "Thoughts?" closers, and personal-story openers HARD-BLOCK. (memory: feedback_no_personal_stories)
• Image-prompt frontmatter (layer_*, metaphor_rationale, image_prompt) is §4a-EXEMPT. Do NOT strip em-dashes there. (memory: feedback_image_prompts_not_4a)
• NEVER reference scarcity hooks ("idle hours", "leftover budget"). Value-first only. (memory: feedback_no_scarcity_offers)

Commands

| ui dashboard | state/skills/skool-partner-cta/resources/ui.openui | |invoke (image): bash state/bin/image/generate-iterate.sh --illustration "<frag>" --text "Help us activate the Partner Network" --slug "partner-cta-$(date +%Y-%m-%d)" --out-dir "/tmp/snappy-image-partner-cta" --format "linkedin-post" --max-auto-turns 1 |verify (image): read /tmp/snappy-image-partner-cta/<slug>/state.json, last frame verdict: "pass" |invoke (voice): import { checkTone } from "../lib/voice.ts" |draft path: state/log/skool-partner-cta/<YYYY-MM-DD>.md |eval log: state/log/evals.ndjson (skill: "skool-partner-cta") |cadence: weekly (cadence_h=168), targets the Wed 1pm ET meetup window

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/skool-partner-cta/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

• Skipping voice.checkTone() for "just a short ask" is the exact path that ships em-dashes
• Closing with Thoughts? to satisfy the question rule still trips the voice gate's engagement-bait check - phrase as genuine curiosity instead
• Round-tripped drafts may show , (space-comma-two-spaces) where em-dashes were stripped; scan the literal string before saving (memory: feedback_skool_emdash_backfill_artifact)
• Fanning out parallel browser writes against one Skool session silently drops uploads - serialize, never parallelize the post step (memory: feedback_no_parallel_browser_writes)
• Drafting without first reading state/log/skool-partner-cta/ for prior drafts produces a same-day duplicate

Self-Test

An agent reading this should correctly:

1. [ ] Stop at the draft step - never auto-post to Skool
2. [ ] Render the image via the validated 6-layer pipeline, not freehand
3. [ ] Include all four Skilljar URLs verbatim
4. [ ] Run voice.checkTone() and refuse to save a body that violates
5. [ ] Pass image-prompt frontmatter through unchanged (em-dashes OK there)

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] skool-partner-cta: <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)] skool-partner-cta: <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.