Tag Taxonomy
Tags are how the librarian routes a task to the right skill. Inconsistent tags mean missed matches; freeform tags mean noise. This page defines the canonical taxonomy every SKILL.md and every entry in registry.yaml must follow.
The five axes
Every skill should carry one tag from each axis that applies. Total tags per skill: hard cap 6.
| # | Axis | Question it answers | Required? |
|---|---|---|---|
| 1 | scope | What kind of skill is this? | Yes — every skill |
| 2 | lifecycle | Which phase of the delivery flow does it belong to? | Yes — every skill |
| 3 | capability | What does it do? | Yes — every skill |
| 4 | domain | Where in the stack does it apply? | Only when domain-specific |
| 5 | risk | How costly is a wrong action? | Yes — every skill |
Pick the smallest closed vocabulary that lets the librarian find the skill. If you cannot map a tag onto one of these axes, drop it.
Axis 1 — scope
What kind of skill this is, in the sense of the four-level hierarchy.
| Tag | Meaning |
|---|---|
core |
Foundation skill — always loaded (00-foundation/). |
category |
Standard skill — loaded on demand (10- through 90-). The default. |
master |
Master skill — orchestrates other skills, no implementation. |
meta |
About skills themselves (templates, taxonomy, generators). |
Axis 2 — lifecycle
Which phase of the work the skill belongs to. The phases form a loop: discovery → design → build → ship → operate → maintain → discovery …
| Tag | What happens here | Categories typically here |
|---|---|---|
discovery |
Understand the problem before solving it. PRDs, brownfield-onboarding, requirement gathering, glossary. | 10-discovery, 25-pragmatism/brownfield-onboarding |
design |
Decide how before doing. ADRs, system design, architecture, planning, decomposition. | 20-architecture, 20-planning |
build |
Make the change. Implementation, refactoring, code generation. | 30-implementation |
ship |
Get the change to production. Tests, review, security, performance, devops, release. | 40-quality, 60-security, 70-devops, 80-collaboration/git-workflow |
operate |
Keep it running. Observability, incident response, debugging, monitoring. | 30-debugging, 70-devops/monitoring |
maintain |
Keep it healthy over time. Upgrades, deprecations, tech-debt tracking, dependency updates, repo cleanup. | 90-maintenance |
Cross-cutting skills (foundation, pragmatism) usually carry no lifecycle tag — they apply everywhere.
Axis 3 — capability
What the skill does. Closed list — extend only by PR.
| Tag | Examples |
|---|---|
auditing |
auditor |
routing |
librarian, orchestrator |
reasoning |
scratchpad, pragya |
principles |
constitution, pragmatism |
efficiency |
token-efficiency |
requirements |
prd, domain-glossary |
planning |
task-decomposition, risk-assessment, dependency-mapping, estimation |
architecture |
system-design, adr, api-spec, data-modeling |
coding |
api-implementation, data-access, code-generation, tdd-workflow |
refactoring |
refactoring, refactoring-suite, cleanup |
resilience |
resilience-patterns, error-handling |
testing |
unit-testing, integration-testing, mutation-testing, testing-strategy |
review |
code-review, pr-management |
security |
secure-coding-review, auth-patterns, dependency-scanning, threat-modelling |
performance |
caching-strategy, query-optimization, profiling, progressive-execution |
docs |
inline-documentation, api-docs, readme-authoring, changelog-generation, release-notes |
collaboration |
git-workflow, pr-management, pair-programming, knowledge-sharing |
devops |
ci-pipeline, docker-containerization, kubernetes-deployment, terraform-iac, monitoring-observability |
debugging |
root-cause-analysis, log-analysis |
maintenance |
repo-maintenance, tech-debt-tracking, dependency-updates, deprecation-strategy, legacy-modernization, version-migration |
pragmatism |
every 25-pragmatism/* skill carries this |
discovery |
brownfield-onboarding, glossary builders |
Aim for one capability tag per skill. Two only when the skill genuinely sits at the boundary (e.g. pr-management is review + collaboration).
Axis 4 — domain
Where in the stack the skill applies. Most skills are stack-agnostic and carry no domain tag. Only add one when the skill genuinely cannot help outside that stack.
| Tag | When to use |
|---|---|
backend |
Server-side-only patterns. |
frontend |
UI-only patterns. |
db |
Database, schema, query, migration patterns. |
infra |
Infra, IaC, container, cluster patterns. |
cross-cutting |
Explicitly applies to all stacks (only if useful as a search term). |
If you find yourself reaching for cross-cutting, you probably do not need a domain tag at all.
Axis 5 — risk
The cost of a wrong action by the agent. Drives mode selection (Eco vs Power) and auditor strictness.
| Tag | Meaning | Examples |
|---|---|---|
advisory |
Skill only emits suggestions; cannot break anything. | dependency-utility-scout, style-conformance, tech-debt-tracking |
reversible |
Changes things, but a single revert restores the previous state. | cleanup, inline-documentation, formatting skills |
irreversible |
Changes that are hard or impossible to undo without data loss or migration. | data-modeling migrations, version-migration, deprecation-strategy |
safety-critical |
Touches security, authentication, production, or release surface. | auth-patterns, secure-coding-review, release-notes, kubernetes-deployment |
If unsure, prefer the higher risk tag — the auditor will be stricter, which is the safe failure mode.
Authoring rules
- Lowercase, hyphen-separated, ≤3 words.
secure-coding-review✅.Secure_Coding_Review✗.secure-coding-review-process✗. - No synonyms. Pick one.
refactor(notrefactoring, notrefactor-skills).db(notdatabase, notdb-schema). - Max ~6 tags per skill. If you need more, your skill is doing too much — split it.
- One per axis. Two in the same axis means the axis is wrong or the skill should be split.
- No "decorative" tags. Every tag must serve discovery.
important,useful,recommended≠ tags. - Closed vocabulary. New tags are added by PR to this file before being used. The
librarianonly knows what is documented here.
Worked examples
# Foundation skill, cross-cutting
- name: pragmatism
tags: [core, principles, pragmatism, advisory]
# axes covered: scope=core, capability=principles+pragmatism, risk=advisory
# lifecycle: cross-cutting → omitted; domain: cross-cutting → omitted
# Category skill in 30-implementation, build phase
- name: refactoring
tags: [category, build, refactoring, reversible]
# axes covered: scope=category, lifecycle=build, capability=refactoring, risk=reversible
# domain: stack-agnostic → omitted
# Category skill in 70-devops, ship phase, infra-specific, dangerous
- name: kubernetes-deployment
tags: [category, ship, devops, infra, safety-critical]
# all five axes used
# Master skill (planned)
- name: release-pipeline
tags: [master, ship, devops, irreversible]
See also
registry.yaml— where these tags actually live.docs/concepts.md— what the four-level hierarchy and Eco/Power modes mean.skills/00-foundation/librarian/SKILL.md— how tags are consumed.
Source: docs/tags.md