Requirements and responsibilities
Assign a named team owner for this task and confirm that all needed service accounts, sources and data are within the approved scope. Never place credentials or private keys in agent prompts.
Use the installed Hermes version and upstream feature documentation for this procedure; it does not establish that xCloud enables the feature by default.
Illustrative situation
Illustrative scenario: a team repeats the same low-risk report cleanup and wants the steps loaded as a named Hermes skill with a test sample.
Choose the approach
Use a profile-local skill for one operator; use a governed shared repository only when an owner can maintain and review shared instructions.
Dashboard and application procedure
Follow these steps yourself, or use the scoped AI handoff below for supported hosting operations.
Step 1 of 5
Select a repeatable bounded task
- Where
- Team workflow owner
- Permissions
- Workflow owner
- Inputs
- Task trigger, required inputs, expected output, exclusions
- Action
- Choose a task with repeatable inputs and a clear completion test; exclude actions involving production changes or external messages.
- Expected result
- The skill’s boundary and acceptance criteria are explicit.
- Verify
- Have a user who performs the task confirm the workflow.
- If it fails
- If the task has no stable inputs or acceptance check, document it as a checklist instead of a skill.
Sources: Hermes skills feature guide
Step 2 of 5
Read current skill guidance
- Where
- Hermes authorized terminal and official skill docs
- Permissions
- Hermes administrator
- Inputs
- Installed Hermes version and official skill format
- Action
- Review the current Hermes skill documentation and identify where skills are stored/loaded for this installation.
- Expected result
- The skill author knows the supported format and loading path.
- Verify
- Check the installed version’s behavior before adding files.
- If it fails
- If the docs differ from the installed release, follow the installed version’s supported interface and record the variance.
Sources: Hermes skills feature guide
Step 3 of 5
Create the supported SKILL.md
- Where
- Hermes authorized terminal or skill_manage tool
- Permissions
- Workspace owner
- Inputs
- Unique skill slug, task trigger/input/output
- Action
- Create `~/.hermes/skills/<slug>/SKILL.md` with YAML frontmatter containing `name: <slug>` and `description: <when to use it>`, then add concise prerequisites, steps, output, acceptance check and stop conditions. This is the documented primary location for agent-created skills.
- Expected result
- The skill describes one task and its limits.
- Verify
- Review for destructive commands, credential exposure and unsupported integrations.
- If it fails
- If any step needs broader access than approved, remove it or seek the appropriate administrator separately.
Sources: Hermes skills feature guide
Step 4 of 5
Load and validate the skill
- Where
- Hermes terminal/CLI session
- Permissions
- Authorized operator
- Inputs
- Created skill slug and a synthetic or approved sample input
- Action
- Start a Hermes session, confirm the new skill is in the loaded skill index, invoke `/<slug>` and run the accepted sample task with no external side effect. If the profile uses `skill_manage` approval, inspect/approve the staged diff from the CLI before relying on the write.
- Expected result
- Hermes recognizes the skill and produces the expected sample output.
- Verify
- Check the actual output against the acceptance criteria and source material.
- If it fails
- If it does not load, verify location/metadata as documented; do not claim installation from file creation alone.
Sources: Hermes skills feature guide
Step 5 of 5
Version and maintain it
- Where
- Team-owned source control or approved workspace process
- Permissions
- Skill owner
- Inputs
- Owner, revision date, test sample
- Action
- Record a maintainer, review date and small regression sample; update instructions when the task or tools change.
- Expected result
- The team can identify the current skill version and owner.
- Verify
- Re-run the sample after changes and compare behavior.
- If it fails
- If the skill becomes stale or unsafe, disable/remove it and return to the manual checklist.
Sources: Hermes skills feature guide
Maintenance
Assign a skill owner and rerun the documented sample after edits or Hermes upgrades; disable the skill if its limits are no longer accurate.
Recovery decisions
If the test fails, disable or revert the skill instructions and return to the manual checklist; do not broaden tools to force completion.
AI handoff
Configure the needed provider, channel, credentials and permissions separately. This guide does not imply xCloud has preconfigured upstream tools or scheduled jobs.
Supported scope
- Review the task result and retain human control app · manual
Human acceptance and follow-up; no external write is implied.
Checkpoint: Verify the source/output and actual side effects before sharing or acting.
Hermes security guide · Hermes daily briefing tutorial · Hermes automate with cron patterns
- Use the authorized Hermes terminal, CLI or documented task interface app · manual
Upstream Hermes CLI/configuration workflow; requires an authorized operator and does not prove xCloud preconfigured the feature or exposed it in its panel.
Checkpoint: Check the installed Hermes version and current upstream instructions; review scope and permissions before running a task.
Hermes quickstart · Hermes cron feature guide · Hermes memory feature guide · Hermes skills feature guide · Hermes agent email address tutorial · Hermes GitHub PR review agent tutorial · Hermes delegation feature guide
Copyable agent brief
Manual checkpoints
- Create under `~/.hermes/skills/<slug>/SKILL.md` using the documented `name` and `description` frontmatter. Confirm the skill is visible and invoke `/<slug>` on a test input.
- Review the task-specific acceptance check and inspect actual output, configuration or run history before relying on it.
Feature coverage
- Documented Hermes task path (covered): Steps identify whether an action uses documented xCloud controls, upstream operator configuration or human review. Select a repeatable bounded task Read current skill guidance Create the supported SKILL.md Load and validate the skill Version and maintain it
- Task-specific acceptance check (covered): The final step validates this guide’s stated result. Version and maintain it