generate-week
Skill0 starsChanged 2 months agoMITView on GitHub
Use when the user asks to plan and generate this week's LinkedIn posts end to end, wants the full weekly pipeline run in one go, or explicitly invokes /generate-week. Plans the week's angles, then generates each post through its own isolated subagent (independent research, draft, critique, visual prompt), checks each against the week's history before moving on, then runs one combined review session and schedules whatever gets approved. Never pads the count to hit a quota.
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- audit-draftSkill
- check-plagiarismSkill
- critique-draftSkill
- critique-draft-substack-articleSkill
- critique-draft-substack-noteSkill
- critique-draft-xSkill
- draft-commentSkill
- draft-replySkill
- extract-hookSkill
- generate-ideasSkill
- generate-visualSkill
- generate-visual-substack-articleSkill
- generate-week-substackSkill
- generate-week-xSkill
- humanize-draftSkill
- interviewerSkill
- monitor-engagementSkill
- optimize-profileSkill
- plan-advocacySkill
- plan-weekSkill
- plan-week-substack-articleSkill
- plan-week-substack-noteSkill
- plan-week-xSkill
- publish-substackSkill
- pull-analyticsSkill
- pull-analytics-xSkill
- repurpose-postSkill
- research-topicSkill
- review-draftsSkill
- schedule-approvedSkill
- schedule-approved-xSkill
- update-playbookSkill
- write-draftSkill
- write-draft-substack-articleSkill
- write-draft-substack-noteSkill
- write-draft-xSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 3 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
Run, in order, via the Skill tool:
- Very specific: line 33
If /plan-week reports any unfilled slot, run one targeted /research-topic <missing-pillar> 2 → /generate-ideas → /plan-week retry, capped …
- Very specific: line 70
Look for another unused status: candidate idea in the same pillar from the existing pool (Post-Ideas/).
- Very specific: line 75
If no usable candidate exists, run /research-topic <pillar> 2 yourself, then /generate-ideas, then pick the strongest resulting idea for …
- Very specific: line 104
After both steps, update Content-Learnings/content-index.md's status column for each row to match its real final state (approved/rejected/ …
- Very specific: line 126
Never skip /review-drafts before /schedule-approved — this skill never approves or schedules anything itself, only sequences the skills …
- Distinctive: line 129
Never let a slot's retry loop (step 2) exceed 2 retries, and never fall back to a weaker or duplicate idea just to fill a day.
- Very specific: line 131
Never pad the week's count.
- Distinctive: line 134
Never attach an image to a scheduled post — no provider is configured.
- Very specific: line 136
If any step's required credential is missing (e.g.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Generate Week (Weekly Orchestrator — Phase 13)
Runs the full pipeline for a week's worth of posts (default target 3, per REQUIREMENTS.md §5) in one invocation, without collapsing them into variations of one template. Plans the week's angles broadly first, then generates each post independently — its own subagent, its own research, its own draft, its own visual concept — while still keeping each post aware of what the rest of the week already covered. Stops for one combined human review session before anything gets scheduled; never approves or schedules anything itself.
Arguments
- week (optional) — passed through to
/plan-week. - target_count (optional) — passed through to
/plan-week. Default 3.
Process
1. Plan the week's angles (broad, not yet deep per-post)
Run, in order, via the Skill tool:
/research-topicwith a broad scan and a count sized for weekly diversity — start at 8, not the skill's own ad-hoc default of 3, sincetarget_countslots need real category spread and some candidates won't survive dedup/verification./generate-ideas(default range)./plan-week [week] [target_count].
If /plan-week reports any unfilled slot, run one targeted
/research-topic <missing-pillar> 2 → /generate-ideas → /plan-week
retry, capped at 2 extra rounds total. After that, accept whatever
/plan-week actually filled — report the honest gap, don't force it.
This step produces status: selected Idea Notes with a target_date each
— real, research-backed ideas (research-topic already verifies against
primary sources before writing a note), not placeholders. What it does
not do is write any post text, critique anything, or think about visuals
— that's step 2, and it happens per post, independently.
2. Generate each post independently, one subagent at a time
For each status: selected idea from step 1, in day order (not in
parallel — each subagent must be able to see what's already been logged
by earlier posts this week), spawn one fresh subagent (Agent tool,
subagent_type: general-purpose, run_in_background: false so this step
stays sequential) with a self-contained prompt that gives it:
- This slot's assigned idea id, topic, pillar, angle, hook, sources, and target date — everything already in its Idea Note.
- The topic + pillar only (not full text) of every other idea assigned this week, so it can keep the week coherent without copying anything.
- The current contents of
Content-Learnings/content-index.md(read it yourself and paste the table in, or point the subagent at the file path — either works) for dedup/history context. - An instruction to Read and follow, directly, step by step:
.claude/skills/write-draft/SKILL.md, then.claude/skills/critique-draft/SKILL.md, then.claude/skills/generate-visual/SKILL.md— applied to its one assigned idea. Do not paraphrase or duplicate those skills' logic into the subagent prompt; point it at the real files so there's one source of truth. - The retry-on-duplicate instruction (this is new, not in the skill
files above): if
/critique-draft's step 2a fires (a genuine duplicate — the draft note gets left atstatus: draftwith aduplicate_flaggedhistory entry instead of advancing toin_review), the subagent must not give up or surface it as-is. Instead:- Look for another unused
status: candidateidea in the same pillar from the existing pool (Post-Ideas/). If one exists, assign it to this slot the same way/plan-weekstep 5 would (status → selected, target_week/target_date set to this slot's), and retry from/write-draft. - If no usable candidate exists, run
/research-topic <pillar> 2yourself, then/generate-ideas, then pick the strongest resulting idea for this slot and retry. - Cap at 2 retries total for this slot. If still stuck after that, stop and report this slot as genuinely unfillable this week — do not force a third attempt or fall back to a weaker/duplicate angle just to fill the day.
- Look for another unused
- An instruction to append or update one row in
Content-Learnings/content-index.mdfor this post: add the row right after/write-draftcreates the Draft Note (id, date, day, status: draft, pillar, topic, angle_summary, hook_gist), then update the same row after/critique-draft(key_examples, conclusion_type, post_format, viral_score, status) and after/generate-visual(visual_format, image_concept). - An instruction to report back concisely: final status (in_review / unfillable), viral_score, visual format + one-line image concept, and whether it had to re-angle (and why).
3. Combined review and scheduling
Once every slot has been processed (some may have ended up genuinely unfilled — that's fine), run:
/review-draftswith no draft-id — it already loops everystatus: in_reviewdraft, so this naturally becomes one combined session covering however many posts made it through the quality gate./schedule-approvedonce, for whatever ended upstatus: approved. Scheduling stays text-only — the existing Buffer mutation has no media field, and no image-generation provider is wired in, so nothing about this changes; the image prompt lives only in each Visual Note.
After both steps, update Content-Learnings/content-index.md's status
column for each row to match its real final state (approved/rejected/
scheduled) — /review-drafts and /schedule-approved don't write to the
index themselves, so mirror their outcomes in here yourself.
4. Final report
- A table: date | day | topic | category | viral_score | visual_format —
covering every slot that made it to
in_reviewor beyond, plus any slot reported unfillable and why. - Week-shape line: one clause per post ("topic 1 → topic 2 → topic 3"),
so you can sanity-check the week's arc at a glance on top of the
per-post
/review-draftssession. - Variety check: how many distinct pillars/formats/hook styles actually appeared across the posts that made it through. If everything technically passed dedup but still reads as samey (e.g. all the same format), say so — this is a flag for you to weigh, not a blocking gate.
- Final counts: how many scheduled, how many approved-but-not-yet-scheduled (shouldn't happen given step 3 runs both), how many rejected, how many slots left unfillable.
Hard rules
- Never skip
/review-draftsbefore/schedule-approved— this skill never approves or schedules anything itself, only sequences the skills that do. - Never let a slot's retry loop (step 2) exceed 2 retries, and never fall back to a weaker or duplicate idea just to fill a day.
- Never pad the week's count. Fewer genuinely distinct, valuable posts than
target_countis the correct outcome when that's what the research supports — report it plainly, don't disguise it. - Never attach an image to a scheduled post — no provider is configured; the deliverable is a prompt for you to run manually.
- If any step's required credential is missing (e.g. Buffer env vars for
/schedule-approved), stop at that step and report exactly what's missing rather than continuing past it or skipping silently.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

