Second Brain

sops-architectural-primitive-between-one-off-and-skill

3 min read

sops-architectural-primitive-between-one-off-and-skill

SOPs — new architectural primitive between one-off tasks and skills

The aj-workspace setup system now has a fourth artifact category: SOPs (Standardized Operating Procedures) at /root/aj-workspace/sops/. SOPs fill the gap between one-off tasks and skills — they orchestrate multi-step operator-driven workflows that chain multiple skills + external tools + human decisions into a single chronologically-ordered, repeatable playbook.

Why SOPs exist (the gap they fill)

Prior taxonomy had three artifacts, none of which could carry a chronologically-ordered multi-skill workflow:

ArtifactWhat it capturesWhat it can’t capture
SkillOne atomic capability Claude can executeChronological orchestration across multiple skills + external tools + operator steps
Topic fileDeep reference knowledge on a single conceptStep-by-step execution order; operator-facing prompts
Feedback fileA single rule + Why + How to applyMulti-step workflow with state handoff
SOP (new)Chronologically-ordered playbook chaining skills + external tools + human decisions—

The triggering case was the Seedance 2.0 website workflow: it calls seedance-loop-prompt → Kie AI (external) → Seedance 2 (external) → website-creator → frontend-intelligence → gh CLI → hostinger skill. Seven artifacts in strict order with parameter handoff and operator decisions in the middle — no single skill could own it; a topic file couldn’t enforce order; only an SOP fits.

Escalation path (new routing rung)

1 occurrence              → just do it, no artifact
2+ operator-driven steps  → SOP (composite workflow)
3+ fully-automatable      → skill (atomic capability)
SOP with automatable core → extract core to a skill; SOP references it by name

Physical structure

/root/aj-workspace/sops/
├── README.md              ← canonical index + conventions + anti-patterns
├── _template.md           ← blank SOP starter
└── <sop-name>.md          ← individual SOPs with flow overview, prerequisites, steps, prompts library, gotchas, source, "Last hardened" date

Framework cascade (completed 11-Apr-2026)

The SOP primitive is wired into 4 locations so no future session can drift:

  1. Topic file: aj-workspace/memory/sops.md — conventions, positioning, SOP-first behavior contract, hardening discipline, Pristine Sweep inclusion
  2. MEMORY.md (aj-workspace) — three edits:
    • Skill & Plugin Gate: added SOP-first sibling rule alongside skill-first
    • Pruning Principle: extended routing tree (2+ operator-driven → SOP, 3+ automatable → skill)
    • Topic Files index: added sops.md pointer
  3. setup-curator SKILL.md — routing decision tree extended: item #8 is now SOPs (multi-step operator-driven workflows), skill promoted to #9 (fully automatable workflows), items 10–12 renumbered
  4. aj-ea MEMORY.md — cross-workspace pointer added to Standing Directives so EA sessions also consult /root/aj-workspace/sops/README.md before starting recurring tasks

SOP-first behavior contract

Before starting any task with SOP shape (multi-step, external tools, manual decisions, non-obvious sequence), Claude MUST:

  1. Check /root/aj-workspace/sops/README.md for a matching SOP (keyword-match on the index or grep the folder)
  2. If found → load it, follow it, use exact skills and prompts it specifies
  3. If not found but task has SOP shape → draft a new SOP from _template.md after the first successful run
  4. Honor gotchas (compounded hard-won learnings)
  5. Update the SOP in-place + bump Last hardened when new gotchas surface

Hardening compounds intelligence

Every successful run hardens the SOP further:

  • New gotchas become rows in the Gotchas table
  • Prompts tighten based on real-world results
  • Run #10 is dramatically better than run #1
  • SOPs that develop overlapping steps get refactored (shared steps → new SOP or skill, Law 1: Zero Redundancy)

Why this benefits all agents (cross-division)

DEV agents (dev-architect, builder, auditor) can now draft SOPs for repeatable infra workflows (MCP server deployment, container migration, credential rotation). EA agents can consult the same SOPs system when facing cross-skill workflows. R&D agents discover workflows that could be elevated from SOP to skill based on automation maturity. The primitive is division-agnostic; the physical home is aj-workspace because that’s where skills, infra, and deployment converge.

Graph View

Backlinks