Skip to content
Back to skills

Routinize

ASecurity

Codify a recurring work pattern into a durable practice (\"관행화\"). Invoke when the same task pattern has recurred and should stop being re-remembered by hand — automatically when you notice a repeat, or explicitly via /routinize. Prime trigger — before-compact time, when the session's accumulated context (the raw material) is about to be compacted away. Follows Greatpractice's capture-eager / promote-deferred discipline (Greatpractice.md §5.4): captures the pattern to memory immediately, but ...

  • 8 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 29, 2026
ai-agentsgo

Security analysis

A100/100

Scanned September 29, 2026

npx -y skills add SoliEstre/EstreGenesis --skill routinize --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Routinize?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Routinize
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/soliestre-routinize/badge)](https://www.skillsdirectory.com/skills/soliestre-routinize)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: routinize
description: "Codify a recurring work pattern into a durable practice (\"관행화\"). Invoke when the same task pattern has recurred and should stop being re-remembered by hand — automatically when you notice a repeat, or explicitly via /routinize. Prime trigger — before-compact time, when the session's accumulated context (the raw material) is about to be compacted away. Follows Greatpractice's capture-eager / promote-deferred discipline (Greatpractice.md §5.4): captures the pattern to memory immediately, but never auto-ratifies a blocking practice — promotion to an enforced entry is user-gated. Skips phronesis-heavy work (judgement-heavy / rare-context) by design."
---

# /routinize — codify a recurring practice (관행화)

Turn "work I keep doing the same way" into a durable Greatpractice entry — without over-codifying. This is the model-invoked front door to Greatpractice's **capture → mature → codify** pipeline (Greatpractice.md §1.2). The full normative model — tiers, maturity score, notability gate, phronesis boundary — lives in `Greatpractice.md`; this skill is the actionable procedure, not a restatement.

**Core discipline — capture is immediate, promotion is deferred** (§5.4). A single session's repetition is usually the *same* trigger coordinate (§5.2), so most invocations end at capture/accumulate; only genuinely independent recurrence produces a draft. The skill never registers an enforcing hook on its own authority — that is a user decision.

## Procedure

1. **Scan for recurrence** — over the recent context, find work patterns that repeated: the same fix applied ≥2×, the same multi-step sequence run again, a "when X, also do Y" coupling, a mistake made → corrected → nearly repeated. Name each candidate pattern in one line.

2. **Phronesis gate — what NOT to codify** (§1.4 / §5.3). Drop any candidate that is *judgement-heavy* (outcome doesn't reduce to one metric), *rare-context* (long-tail work phase), or *high-context-dependent* (the same surface behavior means different things by context). These stay in the agent's instinct — codifying them causes phronesis atrophy. Keep a borderline-but-worth-capturing candidate, but mark `phronesis_boundary: true` (advisory only, never blocking).

3. **Route each surviving candidate** (§5.4):
   - **New pattern** → write `memory/feedback_<slug>.md`: minimal frontmatter (`trigger`, timestamp, `trigger_source`), `lifecycle: probation`, blameless voice (state the pattern, not blame). Eager capture — cost 0, no approval needed.
   - **Already captured** → append this occurrence to the existing feedback's illustration section + record its `(work-domain × phase × time-of-day)` coordinate; bump the significant-coverage counter (substantive appends only).
   - **Notability gate met** — *only* if the pattern now has ≥3 substantive occurrences across **≥2 distinct coordinates** (3× at the same coordinate = 1 trigger, does not qualify) AND a verifiable effect is definable (§5.2) → draft `greatpractice/_propose/<slug>.draft.md` with the full §3 frontmatter, a tier guess (§2: macro governance / mezzo procedure / micro atom — most are mezzo), and the phronesis verdict.

4. **Route ratification to the user — never self-approve** (§5.4 gate 3). A draft becoming an enforced entry — promoted to `greatpractice/{tier}/` with a registered lifecycle hook that will block or warn on future tool calls — is a durable behavioral change. Surface the draft + a one-line recommendation (tier · enforcement_level · why) and let the user decide. If Hyperbrief is adopted and the call is non-trivial, route it through a decision brief.

5. **Report** — 1–3 lines: N captured / N accumulated / N drafted-awaiting-approval / N skipped-as-phronesis. If nothing recurred enough to act on, say so — that is a valid and common outcome (capture-eager does not mean codify-eager).

## Non-goals

- **Not auto-enforcement.** Never registers a blocking hook on its own authority — ratification is user-gated (§5.4 gate 3).
- **Not a summarizer.** It captures *reusable procedure*, not a recap of what happened (that is the compaction summarizer's job — and `/before-compact`'s, which this skill pairs with).
- **Not for one-offs.** A pattern seen once is a memory note at most, not a practice (§5.4 "1 time = raw capture").

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.