Agent Workflows

ldev helps agents because it exposes operational steps that are already useful for humans.

The tool is the execution layer. The value is still in Liferay maintenance work:

• inspect the environment
• diagnose the failure
• apply a fix locally
• verify the result

Context snapshots

Use context first so an agent has the same runtime picture a developer would use.

ldev context --json
ldev status --json
ldev doctor --json

Bootstrap the repo first

Install the managed AI assets into the project:

ldev ai install --target .

Optional overlays:

ldev ai install --target . --project-context
ldev ai install --target . --project --project-context

Useful follow-up:

ldev ai status --target . --json

What this prepares:

• AGENTS.md
• vendor-managed skills under .agents/skills
• optional project context scaffolding
• optional project-owned skills and agents

Real example:

ldev ai install --target . --project --project-context

Project type: ldev-native
Installed skills: 6
AGENTS.md: installed
CLAUDE.md: installed
docs/ai/project-context.md: installed
Updated tool targets: .claude/rules, .cursor/rules, .gemini, .github/instructions, .windsurf/rules, .workspace-rules
Installed project skills: 2
Installed project agents: 4

In Blade workspaces, ldev can coexist with the official AI folders and .workspace-rules model instead of replacing it.

Where knowledge lives

The AI layer is easier to maintain if each kind of knowledge has one clear home:

• .agents/skills/* without the project- prefix are vendor skills installed by ldev
• .agents/skills/project-* are project-owned workflows such as issue handling or PR process
• docs/ai/project-context.md is the long-form project context document
• .workspace-rules/ldev-* are runtime/tooling rules for ldev, not the main home of project process

In short:

• use skills for workflows
• use project-context.md for project context
• use ldev-* workspace rules for runtime and tooling guidance

This keeps the same model working in both ldev-native and blade-workspace without duplicating project process across multiple folders.

Structured portal discovery

Agents can inspect the portal without screen scraping or UI navigation:

ldev portal inventory sites --json
ldev portal inventory pages --site /global --json
ldev portal inventory page --url /home --json

Keeping rules and skills up to date

After pulling a new version of ldev, refresh skills and rules in the project:

ldev ai install --skills-only --target .

This updates .agents/skills/ and all tool-specific rule directories (.claude/rules/, .cursor/rules/, .gemini/, etc.).

Windows: rule directories are created as copies instead of symlinks (symlinks require Developer Mode). Re-running the command above refreshes those copies. If you later enable Developer Mode, the next run replaces the copies with proper symlinks automatically.

Execution, not hype

Use agents for planning or analysis if you want, but keep the system boundary clear:

• ldev discovers state
• ldev executes local operational steps
• ldev returns structured output for verification

That is enough to support reliable agent workflows without inventing a separate platform story.