中文 · Security · License

Start here: Use & develop your own Skill ecosystem

A practical tutorial for using SKILL.md, auditing reusable skills, drafting skills with AI, extracting real workflows, and building a local Skill ecosystem.

npx skills add dongshuyan/compass-skills --skill '*' -a claude-code

COMPASS Skills gives AI coding agents three local skills: task clarification, repo-local task memory, and a local collaboration profile.

The project currently ships three SKILL.md skills:

List the available skills before installing:

npx skills add dongshuyan/compass-skills --list

Install all skills for Claude Code:

npx skills add dongshuyan/compass-skills --skill '*' -a claude-code

Install all skills for both Codex and Claude Code:

npx skills add dongshuyan/compass-skills --skill '*' -a codex -a claude-code

After installation, invoke the skills directly in a session:

$task-clarifier
$task-forest
$user-profile-keeper

For manual installation, copy the three folders under skills/ into the agent's local skills directory and keep their references/, scripts/, and agents/ subdirectories intact.

Long-running agent work needs three kinds of state:

• User context: communication preferences, risk boundaries, recurring omissions, and collaboration style.
• Project context: where the current request fits, what it depends on, and how far it has progressed.
• Goal context: how the current task contributes to the original objective and whether it still matches it.

COMPASS organizes that state into three local workflows:

1. A local profile that the user can inspect and correct.
2. A repo-local task graph that survives session boundaries.
3. A clarification gate before ambiguous or risky execution.

task-clarifier is the entry point for ambiguous, high-cost, high-risk, evidence-sensitive, or externally visible work. It first identifies the user-owned decisions that must be made, asks 1-3 focused questions with recommended answers, confirms shared understanding, and only then searches or executes.

task-forest records long-running work structure: why a task exists, where it fits, how far it progressed, what changed, and what remains unresolved.

user-profile-keeper stores collaboration preferences locally. Future sessions use the profile to ask relevant questions and apply the right risk boundary. Current files, logs, and user-provided context remain the authority; secrets stay out of the profile.

user-profile-keeper -> who is the user and how should we collaborate?
task-forest         -> where does this task fit and is it still aligned?
task-clarifier      -> what should the agent do now?

Task forest HTML export:

Live DAG view:

Task detail view:

User profile and alignment flow:

Ecosystem map:

COMPASS works across agent runtimes as a SKILL.md package with Markdown instructions, YAML frontmatter, optional references/, optional scripts/, and optional agent metadata.

The scripts use Python standard-library components and run locally.

COMPASS keeps runtime data local:

• No upload of task data or user-profile data.
• No browser cookie, token, private key, credential, or session extraction.
• task-forest stores task data under the current workspace, usually .agent-workbench/task-forest/.
• user-profile-keeper stores local profile data under .compass-skills/user-profiles/v1 by default, or a user-selected COMPASS_USER_PROFILE_HOME.
• High-risk actions such as deletion, overwrite, publishing, remote writes, credential use, and global configuration changes require explicit confirmation.

Important: user-profile-keeper uses local plaintext storage without encryption. Do not store passwords, tokens, private keys, verification codes, or highly sensitive personal data in the profile.

See SECURITY.md for the security boundary.

Clarify a task before execution:

Use $task-clarifier to align the task below.

Task: ...
Material: ...
Constraints: ask user-owned decisions first; infer discoverable facts from files, context, or reliable sources. Ask only questions that change scope, method, evidence, format, safety, or acceptance criteria.
Output: ask 1-3 key questions with recommended answers first; once the core need is clear, restate your understanding in 2-5 lines and ask me to confirm.

Maintain the task forest for a workspace:

Use $task-forest to analyze the current session and maintain the task forest for this workspace.

Goal: create a task-forest proposal from long-running goals, tasks, progress, deviations, risks, decisions, and follow-ups in this session.
Requirements:
1. Read the current task-forest list and todo first; initialize task-forest if missing.
2. Identify which long-term goal this session served. If no relation is clear, ask me or create a question/risk node.
3. Save a proposal and show me the planned changes before applying.
4. After approval, apply, validate, export, and report the HTML path.

Initialize a local user profile:

Use $user-profile-keeper to initialize my local user profile.

Goal: build an auditable, correctable, retractable profile from a local questionnaire or the current context.
Boundaries:
1. Store locally only. Do not upload anything or read browser cookies, tokens, or credentials.
2. Do not save secrets, passwords, private keys, verification codes, or browser-session information.
3. Put inferred, private, sensitive, or conflicting claims into pending proposals for my review.
4. Report what was saved, proposed, skipped, or redacted.

The public install path has been validated with skills@1.5.11:

• npx skills add dongshuyan/compass-skills --list finds all 3 skills.
• npx skills add dongshuyan/compass-skills --skill '*' -a claude-code --copy -y installs all 3 skills into a temporary project's .claude/skills/ directory.

Planned additions:

• Build reusable skills from real task histories.
• Upgrade existing skills from observed failures, feedback, and validation evidence.
• Generate restart prompts for long sessions using task-forest as structured context.
• Summarize local agent states, waiting-human items, risks, and review queues.
• Recommend low-switching-cost follow-up tasks from the task graph.

MIT. See LICENSE.

• This repo has been shared as open source on Linux.do.