Usage

Usage#

This guide covers how to use skills from the llm-workflow plugin in your Claude Code sessions.

Invoking Skills#

Skills can be triggered in several ways:

Automatic Triggering#

Skills are designed to trigger automatically when Claude detects a relevant context. Each skill has trigger conditions defined in its description.

For example, the code-review-response skill triggers when you mention:

  • Processing ISSUE comments

  • Code review responses

  • Reviewing code with ISSUE markers

Simply describe what you want to do:

> Process the ISSUE comments in my code

> Find all the code review issues and generate a report

Explicit Invocation#

You can explicitly invoke a skill using the slash command syntax:

> /wf:code-review-response
> /wf:council
> /wf:debate
> /wf:pre-mortem
> /wf:review-tests
> /wf:team-review-tests

This ensures the skill is used regardless of automatic trigger detection.

The debate skill also supports trigger phrases like “debate this”, “argue both sides”, “devil’s advocate”, or “stress test this decision”.

The council skill triggers on “council review”, “multi-judge review”, or “council this plan”.

The pre-mortem skill triggers on “pre-mortem”, “failure analysis”, “risk analysis”, or “what could go wrong”.

The review-tests skill triggers on “review tests”, “audit tests”, “test gaps”, or “test cheating”.

The team-review-tests skill triggers on “team review tests”, “team test review”, “team audit tests”, or “multi-judge test review”.

Skill Output#

Each skill produces specific outputs. See the individual skill documentation for details on what each skill generates.

Common output patterns:

  • Reports - Markdown files in the skill-specific resources/ subdirectories

  • Transcripts - Debate transcripts in resources/agent-docs/reviews/debates/

  • Council reports - Verdict reports in resources/agent-docs/reviews/councils/

  • Pre-mortem reports - Risk registers in resources/agent-docs/reviews/pre-mortems/

  • Test review reports - Findings registers in resources/agent-docs/reviews/tests/

  • Modified files - Direct edits to source files

  • Console output - Information displayed during execution

Default Input Files#

Generated skill artifacts use platform-specific default input files:

  • Claude Code: .claude/<skill>.md

  • Codex: .codex/<skill>.md

  • OpenCode: .opencode/<skill>.md

  • Grok: .grok/<skill>.md

Migrated skills also check the legacy .claude/<skill>.md path after the platform default, so existing repositories keep working.

Working with Reports#

Several skills generate markdown reports for later processing. These reports are designed to integrate with other tools:

  • Spec generation - Reports can feed to-spec / to-slices in the slice loop

  • Documentation - Reports serve as project documentation

Example Workflow#

A typical code review workflow:

  1. Review code manually - Add ISSUE: comments where you find problems

    # ISSUE: This function lacks error handling for empty input
def process_data(items):
    return [transform(i) for i in items]

  2. Generate report - Ask Claude to process the issues

    > Process the ISSUE comments in my codebase

  3. Review the report - Check resources/agent-docs/reviews/code/code-review-1.md

  4. Plan fixes - Use the report to create a PRD or task list

  5. Implement - Fix the validated issues

Tips#

  • Be specific - When invoking skills, provide context about what you want

  • Check outputs - Review generated reports before acting on them

  • Iterate - Skills can be run multiple times as you refine your approach

Contents