Capabilities
The building blocks behind every AI workflow. This page explains what each capability type does and how they compose during real use.
Capability Types
| Type | What it is | How you use it |
|---|---|---|
| Skills | Domain expertise with triggers, protocol, and constraints | Activated automatically when your request matches |
| Commands | Shortcuts that invoke a specific skill or workflow | Type directly: /gm, /send, /consolidate |
| Extensions | Runtime integrations for scheduling and live platform access | Used by skills behind the scenes |
| MCP Servers | External tools that provide structured data access such as vault search | Called by skills behind the scenes |
How They Fit Together
When you interact with the system, these layers compose:
You type a request or command
→ the assistant matches it to one or more skills
→ skills call extensions or MCP servers as needed
→ results flow back through the skill to you
Example: morning briefing
- You type
/gm - The
/gmcommand invokes the briefing skill - The briefing skill calls the QMD MCP server to search the vault for goals
- The briefing workflow uses the glab skill to check GitLab for open MRs and pipeline failures
- It synthesizes a priority-ranked briefing and writes it to
$OSDU_BRAIN/00-inbox/(or prints to stdout if the vault is not configured)
The key distinction: commands and natural-language requests are what you type. Skills, extensions, and MCP servers are what the system uses to fulfill them. Skills are the primary execution unit; commands invoke workflows, and extensions and MCP servers provide external access.
Skills
Skills are the primary execution unit. Each is a SKILL.md file containing trigger rules, a step-by-step protocol, constraints, and an expected output format. When you say something that matches a skill's triggers, it activates automatically.
All skills live in skills/ and are available to every context. Specialist agents use the shared skills plus their own domain knowledge.
Observe and Plan
| Skill | What it does | Common triggers |
|---|---|---|
| brain | Read/write the Obsidian vault: briefings, reports, decisions | vault, brain, decisions, remember |
| glab | GitLab CLI operations: MR/issue workflows, pagination | GitLab, MRs, pipelines |
| health | Environment health: cluster state, deployed services, safeguards | cluster status, what's deployed |
| osdu-activity | OSDU platform monitoring: MRs, pipelines, issues across 30+ repos | pipeline status, MR activity |
| osdu-engagement | Team contribution analysis: commits, reviews, ADR participation | contributions, reviews |
| osdu-quality | Test reliability analysis: pass rates, flaky tests, provider metrics, acceptance test parity | test reliability, flaky tests, coverage gaps |
| learn | Knowledge acquisition from external sources (wikis, docs, URLs) | learn, study, index knowledge |
| briefing | Daily briefing: GitLab MRs + vault goals | gm, good morning, briefing |
| consolidate | Vault hygiene: stale notes, contradictions, decay | consolidate, stale notes |
| setup | Environment validation: tools, auth, extensions | setup, onboarding |
Ship and Automate
| Skill | What it does | Common triggers |
|---|---|---|
| send | Review-commit-push-MR workflow (worktrunk or git) | ship, send |
| loop | In-session recurring tasks via Cron extension (session-scoped) | loop, repeat, every X minutes |
Operate
| Skill | What it does | Common triggers |
|---|---|---|
| iac | Azure IaC: Terraform patterns, AVM modules, Helm/Kustomize, safeguards, debugging, verification | Terraform, AVM, AKS, debug, verify |
| clone | Clone OSDU GitLab repos (worktree or standard git) | clone, set up repo |
| osdu-data-load | Load OSDU datasets (reference data, TNO, Volve, NOPIMS) into any instance | load data, bootstrap instance, what datasets |
| osdu-qa | Route QA commands to appropriate sub-agents | test, qa, run tests, check |
| dependencies | Dependency analysis: Trivy scanning, POM analysis, risk scoring | dependency scan, CVEs |
| dependency-scan | Full project dependency analysis with vault reports | scan dependencies, full analysis |
| maven | Maven version checking and security scanning | Maven versions, security scanning |
| acceptance-test | Run Java acceptance/integration tests from service repos against live cimpl environments | acceptance test, integration test, test service |
| build-runner | Execute build/test commands with structured output | build, test, Maven verify |
| fossa | Fix FOSSA NOTICE file from failed MR pipeline | fossa, notice file |
| maintainer | GitLab MR trusted branch sync for maintainers | maintainer, allow, sync branch |
| remediate | Execute dependency remediation from analysis report | remediate, update dependencies |
Commands
What you type directly. These are the user-facing surface of the system.
| Command | What it does |
|---|---|
/gm | Morning briefing: gathers GitLab activity + vault goals |
/briefing | Daily briefing (same as /gm) |
/send | Ship local changes: review, quality checks, commit, push, MR |
/brain | Read from or write to the Obsidian vault |
/consolidate | Scan vault for stale knowledge and contradictions |
| Command | What it does |
|---|---|
/debug | Investigate an infrastructure issue with systematic RCA |
/verify | Verify infrastructure work is complete with evidence |
| Command | What it does |
|---|---|
/clone | Clone OSDU repositories into the workspace |
/dep-scan | Run dependency analysis with risk scoring |
/fossa | Fix a FOSSA NOTICE file from a failed MR pipeline |
/maintainer | Maintainer actions for OSDU MRs: review or allow |
/remediate | Execute dependency remediation from an analysis report |
Extensions and MCP Servers
These integrations extend what the system can access. Skills call them behind the scenes; you rarely interact with them directly.
In-session job scheduling: cron expressions, intervals, one-shot timers. Tasks are session-scoped and vanish when you exit.
| Tool | Purpose |
|---|---|
cron_create | Register task with cron expression + prompt |
cron_list | List all tasks with next-run times |
cron_delete | Remove task by ID |
cron_engine_status | Task count and registry mode |
Used by the loop skill for recurring tasks like "check pipeline status every 10 minutes." Maximum 50 tasks, auto-expire after 3 days.
Live OSDU platform access via Keycloak client credentials. 15 tools across 6 services, read-only by default.
| Service | Tools | What you can query |
|---|---|---|
| Health | 1 | Connectivity, auth, service probes |
| Entitlements | 1 | User groups |
| Search | 3 | Query, find by ID, list by kind |
| Storage | 5 | Get/list/fetch records, versions |
| Schema | 2 | List and get definitions |
| Legal | 2 | List and get legal tags |
Write and delete operations require explicit opt-in via environment variables. Used by the @osdu agent and health skill for live platform queries.
Hybrid search (BM25 + vector + cross-encoder reranking) over the Obsidian vault. Optional; available on macOS and Linux only.
| Tool | Purpose |
|---|---|
qmd-query | Hybrid search with lex, vec, or hyde sub-queries |
qmd-get | Retrieve a specific note by file path or docid |
qmd-multi_get | Retrieve multiple notes by glob pattern |
qmd-status | Check vault indexing status |
Used by the brain, briefing, learn, and consolidate skills. See Brain for vault details.
See also
- Specialist Agents — which agents use which skills and how delegation works
- Brain — how the
brainandlearnskills interact with persistent memory - Developer Workflows — end-to-end patterns built from these capabilities