what-should-i-automate
Skill0 starsChanged 4 months agoMITView on GitHub
On-demand automation-gap audit — surfaces repetitive manual toil worth killing, weighted by LIVED pain-signal (frequency × friction from real recent work + conversation), not a static config scan. PULL-only: Nile invokes it when he has build-energy. Triggers — "/what-should-i-automate", "what should I automate", "what's worth automating", "what toil can I kill", "automation gaps". Anti-guilt by design (opportunities for when you have energy, NOT a backlog of shoulds), respects intentional-manual (flags Chesterton's-Fence cases, never asserts), and routes accepted candidates to skill-creator / Todoist / scheduled-tasks-manage rather than auto-building. NEVER schedule this skill — if asked to cron it, refuse and explain (a pushed "here's more you haven't done" report is a guilt-generator). Complements the always-on skill-candidate-diagonal + Shadow-Offload rules; this is the deliberate deep pass. On-demand only — never auto-fire.
In this repository9 other files it gives its agents.
- adhd-friendlySkill
- atom-cypher-safeSkill
- document-mind-mapSkill
- handoffSkill
- interview-modeSkill
- notebooklm-bulk-labelSkill
- prompt-forgeSkill
- skill-scoutSkill
- zero-as-answerSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
If nothing clears the bar, say so plainly ("nothing worth the spoons right now") — a clean empty result is a valid, honest answer, not a …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
what-should-i-automate — On-Demand Toil Audit (PULL-only)
A deliberate hunt for repetitive manual work worth automating — run only when Nile asks, because he has the build-energy in that moment. This skill exists in hypothesis-tier (10×10× c_r≈0.42, 2026-06-15): the idea is sound, but it survives only as a pull-tool with the four guardrails below. Violate any one and it becomes net-negative.
The four hard constraints (these ARE the skill — from the 10×10× kills)
- PULL, never PUSH. Never scheduled, never a cron, never volunteered. If Nile (or any config) asks to schedule/cron/automate the running of this skill → refuse and explain: a monthly "here's more you haven't done" report is a shoulds/guilt generator for a burnt-out, backlog-heavy, no-new-habits regime. The whole point is he runs it when he wants to build.
- Pain-signal first, config-scan last. The best gap-signal is lived friction — which task annoyed him, how often he did it by hand — visible only in real work, not a static file scan. Weight conversational/session/Todoist friction above infra-config toil.
- Anti-guilt framing. Output = a SHORT menu of opportunities, framed as "when you have energy, this kills X pain" — never "you should have automated this." Honor [[feedback_dont_underestimate_capacity]] + [[feedback_shame_guilt_fear_watch]]. If burnt-out/low-Mana signals are present, return fewer candidates, lighter touch — or just the single highest-pain one.
- Respect intentional-manual (Chesterton's Fence). Some things are manual on purpose (e.g. YNAB new-txn approval = Nile's deliberate control point; irreversible/financial/judgment ops). For any candidate that might be deliberate, flag it as "keep-manual?" and ask — never assert "automate this." Cross-ref [[feedback_always_ask_list]] + Chesterton's Fence.
Signal sources (pain-weighted, high → low)
- Lived friction (highest weight) — repetitive manual steps in this session + recent conversation: what did Nile do by hand more than once? what did he sigh at? Pull from session context + the capture buffer (
_PARA/_Inbox/claude-session-capture.md). - Recurring task patterns — Todoist tasks he keeps re-creating, or that recur manually (the Shadow-Offload "3+ times" signal).
mcp__todoist__*recent/recurring. - Workflow/infra toil (lowest) — manual multi-step sequences in scripts/ops that a wrapper would collapse. Lightweight; the
config-drift-scansurface. Don't lead with this — it's the static-scan trap.
Scoring & output
Per candidate: frequency × friction × automatability − intentional-manual penalty. Rank, then return Nile's recommendation format:
- Top 3–5 (cap small — anti-overwhelm), ranked. For each: the toil it kills · how often it bites (the pain) · Spoon Cost (1–5) to automate · First Brick (smallest start) · a "keep-manual?" flag where Chesterton applies.
- If nothing clears the bar, say so plainly ("nothing worth the spoons right now") — a clean empty result is a valid, honest answer, not a failure to find work.
Routing accepted candidates (surface → Nile picks; never auto-build)
- Repeatable procedure (workflow with steps) →
skill-creator(the skill-candidate-diagonal). - One-off action → Todoist task (action-verb-first).
- Genuinely periodic + low-guilt (e.g. a backup check) →
scheduled-tasks-manage(notify-on-fail wrapper). Apply the cron-vs-pull judgment honestly — most "automations" are better as pull-tools too.
Boundary
The always-on skill-candidate-diagonal + Shadow Offload rules already surface gaps passively during work (this is correct and should keep happening). This skill is the deliberate, on-demand deep pass for when Nile sets aside energy to hunt toil — it does not replace the passive rules, and it must add the pain-weighted ranking they don't.
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.

