Help

FM-Agent help

Answer in the user's language. Explain only; do not start analysis, change configuration, create a baseline, or modify project files. Direct a request for current run state or a particular finding to diagnose instead.

Explain the available skills

State that the plugin exposes these five skills:

| Skill | Use it for | | --- | --- | | run | Analyze a Git project and automatically select full, incremental, or no-op. | | install | Check prerequisites and explain missing setup. | | config | Save non-secret defaults such as scope, retries, knowledge files, and supplemental edges. | | diagnose | Read the latest run state, summary, or a stored bug report without starting another analysis. | | help | Explain workflow, usage, boundaries, and current capabilities. |

Do not promise a particular slash-command spelling: Codex and Claude Code may present the same shared skills through different user interfaces. Give natural language examples such as “analyze the current Git project with FM-Agent” or “configure FM-Agent to analyze only backend/”.

Explain automatic mode selection

Explain that users do not choose a low-level full or incremental flag. The run skill deterministically inspects the baseline and chooses:

| Condition | Mode | | --- | --- | | No compatible successful baseline | Full analysis | | Compatible baseline and no supported source-file change | No-op | | Compatible baseline and supported source-file change | Incremental analysis |

A baseline is reusable only when its scope and configuration fingerprint, function artifacts, specifications, verification results, and Git commit remain compatible. A scope, knowledge-file, or supplemental-edge change deliberately causes full analysis.

A source-file change means an added, modified, or deleted file whose extension is supported by the plugin. Documentation-only changes do not start an incremental analysis.

Explain output and boundaries

Explain the artifact boundary:

• fm_agent/ contains FM-Agent-compatible analysis artifacts: phases, extracted

function copies, top-down layers, verification results, and any bug-validation output.

• fm_agent_plugin/ contains plugin control state: saved configuration, locks,

run records, baseline fingerprints, analysis indexes, graph precision, incremental decisions, and isolated probe builds.

• Project business source is never annotated with [SPEC] or [INFO].

Explain that the same skills/, scripts/, and references/ are shared by Codex and Claude Code. Their manifests only provide discovery; each skill resolves its plugin root from its own SKILL.md location as defined in [runtime-path.md](../../references/runtime-path.md).

Explain that run detects optional CodeGraph support before analysis. When it is available, full and incremental analysis automatically remove and regenerate the derived CodeGraph index without requesting authorization. The plugin does not install unavailable software; it records a best-effort static call graph rather than claiming exact graph precision.

Explain the current capabilities

Currently, the plugin supports a full run_pipeline and an automatic baseline-driven analysis workflow.

Scripts coordinate plugin state and validate artifacts.

The coding agent performs repository understanding, specification writing, and reasoning during the run workflow.

Describe and execute only capabilities documented by this plugin's shared skills and references. When uncertain, state the documented current behavior rather than inferring features from the original FM-Agent project.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fmagent-project
• Source: fmagent-project/FM-Agent-Skill
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.