Skip to content

Agent Skills#

Agent Skills teach AI agents how to guide users through terraform-ingest workflows. Skills live in ./skills/<skill-name>/SKILL.md using the Claude Code skill format and deploy to any supported agent via the CLI.

Overview#

Skills complement the MCP service by covering setup and orchestration before an agent connects to MCP, and by providing CLI fallbacks when MCP is unavailable.

Skill Purpose
terraform-ingest Hub — routes to the right workflow
terraform-ingest-setup Install, init config, git credentials
terraform-ingest-configure Build and validate config.yaml
terraform-ingest-ingest Run ingestion and verify output
terraform-ingest-release-audit Pre-push QA gate before hand-off

Phase 2 (planned): terraform-ingest-mcp-connect, terraform-ingest-query
Phase 3 (planned): terraform-ingest-import, terraform-ingest-troubleshoot

CLI management#

List, validate, and deploy skills from the application:

terraform-ingest skills list
terraform-ingest skills show terraform-ingest
terraform-ingest skills validate
terraform-ingest skills validate-config config.yaml
terraform-ingest skills install --scope project --target cursor
terraform-ingest skills install --scope user --target claude_code --dry-run

Supported deploy targets: cursor, claude_code, codex, opencode, github_copilot, windsurf, hermes, openclaw.

When --target is omitted, install auto-detects agents from existing config directories.

First-time user flow#

An agent should run these skills in order:

  1. Setup — uv sync, terraform-ingest init config.yaml
  2. Configure — add repositories via terraform-ingest config add-repo
  3. Validate — terraform-ingest skills validate-config config.yaml
  4. Ingest — terraform-ingest ingest config.yaml
  5. MCP (optional) — connect agent per MCP guide

Config validation#

terraform-ingest skills validate-config config.yaml

Checks:

  • YAML parses and matches IngestConfig schema
  • output_dir and clone_dir are writable
  • Warns if repositories is empty

Interface selection#

Agents should prefer MCP tools when connected, otherwise use CLI:

Task MCP CLI
Start ingest run_ingestion terraform-ingest ingest config.yaml
Poll progress get_ingestion_status terraform-ingest ingestion status
Add repo — terraform-ingest config add-repo
Deploy skills — terraform-ingest skills install

Release readiness#

During iteration:

task qa:quick

Before hand-off or push:

task qa:prepush

Read .terraform-ingest/summary.json for pass/fail — do not paste log files into agent chat. See QA Pre-Push Gate.

Full design document: docs/superpowers/specs/2026-06-10-terraform-ingest-agent-skills-design.md

Using with Cursor / Claude Code / Codex#

Deploy project-local skills so your agent discovers them automatically:

terraform-ingest skills install --scope project --target cursor
terraform-ingest skills install --scope project --target claude_code

Using with other agents#

Agents that read project files can load skills/<name>/SKILL.md directly. MCP-connected agents should use MCP tools for query and runtime ingest after the setup skills complete.