Skip to main content
Skills are reusable instruction packages for specialized work. Each package has a SKILL.md file with a name, a description, and the workflow the agent should follow. A package can also include reference files. Embedder ships with skills for embedded platforms, hardware debugging, dashboards, plotting, optimization, and static-analysis workflows. You can install more skills for one project or for every project on your machine.

​
How Embedder discovers skills

Embedder scans recursively for files named SKILL.md. It combines the bundled catalog with skills from these locations:
  • User locations: ~/.embedder/skills, ~/.claude/skills, and ~/.agents/skills
  • Project locations: .embedder/skills, .claude/skills, and .agents/skills
A filesystem skill with the same frontmatter name as a bundled skill replaces the bundled definition in the catalog. Platform-scoped skills are hidden when the selected platform does not match their metadata. At the start of a conversation, Embedder takes a snapshot of the available skill names, descriptions, and locations. It watches skill directories for later changes. If you install a new skill during the conversation, Embedder announces it separately and can load it immediately. A new conversation includes it in the regular skill catalog.

​
Invoke a skill

You can invoke a skill in three ways:
  1. Ask for work that matches its description. The agent should load the matching skill before it continues.
  2. Name the skill in your request, such as Use debug-gdb to inspect this fault.
  3. Run its generated slash command, such as /debug-gdb inspect the hard fault, when the skill name is a valid command name and does not conflict with a built-in command.
When the agent invokes a skill, Embedder loads SKILL.md and its supported companion files into the conversation context. Skill lookup by the agent is case-insensitive, but using the displayed name avoids ambiguity.
Loading a skill gives the agent instructions. It does not grant permissions, install a toolchain, connect hardware, or bypass the current mode.

​
Automatic loading

A skill can declare autoLoad conditions in its frontmatter:
Embedder recognizes these conditions:
  • always: load the skill whenever its other conditions match.
  • manufacturers: match the manufacturer of a selected project component.
  • platforms: match a normalized platform name exactly.
  • platformPrefixes: match the start of a normalized platform name.
  • hardwareProviders: match connected hardware, such as gdb, serial, or bluetooth, in Debug mode.
  • modes: restrict loading to named modes such as act or debug.
  • compliance: match the project’s selected per-file compliance standard or standard family.
Platform and manufacturer conditions also limit whether you can invoke the skill directly. For example, a board pinout skill is unavailable when another platform is selected.

​
Install a skill

Use the built-in installer for local packages and remote HTTPS sources:
The default destination is the current project’s .embedder/skills directory. Add --user only when you want the skill in every project:
Other options are:
  • --overwrite: replace a skill with the same name. The installer never replaces one by default.
  • --project-root "<path>": install into a project other than the current working directory.
The installer requires a top-level SKILL.md. Its frontmatter name must contain lowercase letters, numbers, and single hyphens between segments. Installation is atomic, so a failed replacement leaves the existing skill intact. For remote packages, the installer accepts raw HTTPS files, GitHub blob URLs, and GitHub tree URLs. It rejects embedded credentials, non-HTTPS redirects, path traversal, and symbolic links. A package can contain at most 100 files and 1 MB total, with a 250 KB limit per file.
Read a remote skill before installing it. A skill can tell the agent to edit files or run commands within your existing permission model.

​
Bundled skill catalog

Bundled skills do not carry a formal category field. The shipped catalog falls into the groups below. Your selected platform, connected hardware, mode, and compliance profile determine which entries you see.

​
Skill management

  • install-skills provides the guarded installation workflow and loads automatically in Act and Debug modes.

​
Vendor SDKs and project workflows

  • STMicroelectronics: STM32, STM32N6, and stm32-cubemx2
  • Infineon: infineon-aurix and infineon-aurix-cli-build
  • NXP: nxp and nxp-imxrt1186
  • Nordic Semiconductor: nrf-zephyr
  • Renesas: renesas-ra-fsp
  • Silicon Labs: silabs-simplicity-studio and sixg301
  • XMOS: xmos
  • Board-specific operational workflows: psoc-edge-e84 and APOLLO510_EVB

​
Board and pinout references

  • ST Nucleo board and PCB-group skills use NUCLEO_* names and cover the bundled Nucleo-32, Nucleo-64, Nucleo-64-P, and Nucleo-144 families.
  • Infineon board skills use KIT_* and TRIBOARD_* names. The bundle also includes AURIX_TC4X_MOTOR_POWER_BOARD.
  • Other board references include renesas-ek-ra8p1, nxp-imxrt1186, psoc-edge-e84, sixg301, and APOLLO510_EVB.
The exact Infineon board skill IDs are below. Skill IDs and catalog platform IDs are not always one-to-one: KIT_A2G_TC397_TFT is the shared skill for the catalog platforms KIT_A2G_TC397_3V3_TFT and KIT_A2G_TC397_5V_TFT. Its auto-load aliases include both underscore and hyphen spellings for each board.
The exact ST board and PCB-group skill IDs are:
See Supported hardware for every board model covered by each grouped skill.

​
Debuggers, transports, and instruments

  • Debuggers and trace: debug-gdb, debug-jtrace, debug-itm, and debug-rtt
  • Serial and wireless: debug-serial and debug-bluetooth
  • Logic and mixed-signal instruments: debug-saleae, debug-digilent, debug-picoscope, and debug-oscilloscope
  • Power instruments: debug-joulescope and debug-ppk2
Most hardware-provider skills load in Debug mode when the corresponding provider is connected.

​
Telemetry and dashboards

  • plot covers live Teleplot streams, plot configuration, XY views, and derived math channels.
  • dashboard-authoring covers project-local Monitor dashboards and the built-in component API.
  • dashboard-custom-components covers SVG and Canvas components when the built-in API does not provide the required view.

​
Optimization and protocol analysis

  • optimize covers measured firmware optimization workflows.
  • can-codebase-decoder derives CAN decoding context from the current codebase and can load for serial debugging.

​
Coding standards and validation

  • misra-c-2012-coding supplies coding guidance when a matching compliance profile is active.
  • validate-CERT and validate-MISRA define evidence-based validation workflows.

​
Coverity

  • coverity-cli, coverity-connect, and coverity-compliance cover local analysis, Connect workflows, and compliance configuration.
  • validate-coverity checks the resulting analysis evidence.

​
Parasoft C/C++test

  • parasoft-cpptest-cli and parasoft-cpptest-mcp cover CLI and MCP workflows.
  • validate-parasoft checks configuration, execution, and report evidence.

​
Perforce QAC

The QAC bundle includes:
  • Setup and interfaces: qac-install, qac-quick-start, qac-cli, qac-eclipse, qac-vscode, and qac-live
  • Core analyzers and modules: qac-qac, qac-qacpp, qac-qarust, qac-dataflow, qac-mta, qac-rcma, and qac-namecheck
  • Compliance modules: qac-ascm, qac-barrccm, qac-certccm, qac-certcppcm, qac-cweccm, qac-cwecppcm, qac-hkccm, qac-hkcppcm, qac-m2cpp, and qac-mc25cm
  • Product references and integration: qac-manual, qac-release-notes, and qac-validate
  • End-to-end evidence checking: validate-qac

​
Create a compatible package

A minimal skill needs name and description frontmatter:
Put references beside SKILL.md when the workflow needs fixed templates or concise supporting material. Use the installer after authoring so Embedder validates the package before placing it in a discovery directory.
Last modified on August 24, 2026