i-have-adhd: ADHD-friendly concise-output skill for coding assistants
Provides coding assistants an action-first, numbered ADHD-friendly output style to cut verbosity and clarify next steps for faster execution.
GitHub ayghri/i-have-adhd Updated 2026-07-22 Branch main Stars 6.9K Forks 283
TypeScript/Node.js LLM plugin Output-style optimization ADHD-friendly

💡 Deep Analysis

4
What specific problem does this project solve, and what is its value proposition?

Core Analysis

Project Positioning: This project codifies an “action-first, stepwise, next-step and time-estimate” output style as an installable skill that constrains coding assistant responses to be directly actionable, reducing friction from recommendation to execution.

Technical Features

  • Rule-driven: Output style is explicitly defined by 10 rules in SKILL.md, allowing review and governance.
  • Plugin-based: Installs via the host agent (Claude Code, Codex) plugin mechanisms and requires no separate backend runtime.
  • Reusable/Customizable: Forking and editing skills/i-have-adhd/SKILL.md lets teams create tailored interaction policies.

Usage Recommendations

  1. Use for engineering tasks: Best for bug fixes, small code changes, debugging, and ops steps where immediate actions are needed.
  2. Calibrate rules: Teams should adjust time estimate granularity and “visible wins” formatting to match team velocity expectations.
  3. Regression test: Validate on representative prompts before rolling to other agents.

Cautions

  • Context omission risk: Strict concision can omit required assumptions; for complex tasks, request/attach extra context.
  • Compatibility dependence: Effectiveness depends on host agent’s skill/plugin semantics.
  • Privacy sensitivity: The rule to “restate state every turn” can inadvertently expose secrets; add protective constraints to SKILL.md.

Important Notice: Prioritizing executability trades off teaching depth and exploratory analysis—use this as an execution-focused tool, not a comprehensive teaching assistant.

Summary: Valuable for making assistant output directly actionable in engineering workflows, but requires customization and testing to avoid lost context and privacy issues.

90.0%
What is the learning curve for end users and maintainers to install and use this skill? What common issues arise and what are best practices?

Core Analysis

Key Question: What is the learning curve for end users and maintainers? What common issues occur and what are practical best practices?

Technical Analysis

  • End users: Installation/calling is low friction (e.g., claude plugin marketplace add ..., then /i-have-adhd). The output style is immediate and reduces cognitive load.
  • Maintainers/customizers: Need to understand SKILL.md, the host agent’s plugin model, and regression testing across scenarios—moderate learning curve.
  • Common issues:
  • Over-simplification leading to missing assumptions.
  • Host agents varying in support for plugin capabilities, so rules may be treated as suggestions.
  • Inaccurate minute-level time estimates affecting scheduling.

Practical Recommendations (Best Practices)

  1. Pilot on a small set: Validate with 5–10 representative prompts before team-wide rollout.
  2. Govern changes: Put SKILL.md edits through PRs and maintain regression tests.
  3. Sensitive-data exemptions: Add clauses preventing automatic restatement of credentials or secrets.
  4. Calibrate time estimates: Use historical team data to round estimates (e.g., to 5-minute buckets).

Cautions

  • Don’t skip regression testing: Host-dependent behavior requires validation in the target environment.
  • Provide expansion paths for complex tasks: Design an explicit “request context” step when rules would cause missing information.

Important Notice: End users see fast, tangible benefits; maintainers must invest in tests, customization, and governance to ensure reliability and safety.

Summary: Low barrier to value, but governance and engineering work are needed to control inaccuracies and privacy exposure.

88.0%
How should teams safely and effectively customize and maintain SKILL.md (including privacy and regression testing strategies)?

Core Analysis

Key Question: How should teams safely and effectively customize and maintain SKILL.md to balance executability, privacy, and consistency?

Technical Analysis

  • Governance (VC + review): Keep SKILL.md under repo control and require PR reviews for changes, documenting rationale and tests.
  • Regression test suite: Maintain representative prompts and expected outputs; automate checks for numbered steps, presence of next-step, and proper time-estimate formats.
  • Privacy protection rules: Add explicit exemptions in the rule file (e.g., disallow restating fields containing secret, token, password) and educate users in the install docs.
  • Cross-agent compatibility validation: Run regression suites on target agents (Claude Code, Codex), log divergences, and note downgrade strategies.

Practical Recommendations (Steps)

  1. Init governance template: Add CONTRIBUTING.md describing change and testing requirements.
  2. Build regression corpus: At least 20 representative cases covering fixes, debugging, ops; automate structural checks.
  3. Add privacy static checks: CI scans that block examples or rule edits that echo credentials.
  4. Monitor estimate calibration: Collect estimate vs actual time delta and adjust rules periodically.
  5. Define downgrade rule sets: Create a strict-subset of rules for agents that lack post-processing support.

Cautions

  • Do not hardcode restatement of secrets: Limit the state-restatement rule to non-sensitive fields or to redacted summaries.
  • Iterate from failures: Use failing cases as primary inputs to revise SKILL.md.

Important Notice: Treat SKILL.md as a governed product artifact—tests, PR reviews, and privacy controls are required when deploying across teams.

Summary: Governance + regression tests + privacy scans + cross-agent validation preserve the skill’s execution benefits while keeping risk manageable.

87.0%
Technically, how does the skill enforce or guide the assistant to follow SKILL.md rules? What is the architecture and integration flow?

Core Analysis

Key Question: How does the skill convert the 10 rules in SKILL.md from a text asset into enforceable assistant behavior?

Technical Analysis

  • Install and Trigger Path: README indicates the repo is fetched by the host agent and the style is activated via explicit commands (/i-have-adhd or $i-have-adhd); some agents may implicitly trigger when a task matches.
  • Likely Implementation Mechanisms:
  • System or tool prompt injection: The agent injects rules as system-level or request-level instructions before generation.
  • Template-constrained outputs: Formatting constraints (numbered steps, time estimates) are expressed as templates.
  • Post-generation hooks/validation: If the host supports it, the agent can validate and rewrite nonconforming output.
  • Architectural trade-offs: Lightweight and backend-free; ease of install and update. Enforcement strength depends on the host agent’s ability to accept system prompts or perform post-processing.

Practical Recommendations

  1. Verify host capabilities: Ensure the target agent supports injecting system prompts or post-generation hooks via plugins.
  2. Regression test with examples: Use representative prompts to verify stable adherence to numbering, next-step, and time estimates.
  3. Prepare fallbacks: If the host lacks post-processing, tighten generation instructions in SKILL.md to increase compliance.

Cautions

  • Do not assume absolute enforcement: On some hosts, rules may act as recommendations and the model may still stray.
  • Testing is essential: Behavior is host-dependent and must be validated in the target environment.

Important Notice: This is a lightweight constraint mechanism—valuable for improving outputs but not guaranteed to fully enforce rules across all hosts.

Summary: The path is rule text -> plugin/system prompt or post-processing -> generated output. Success hinges on the host plugin API and hook support.

86.0%

✨ Highlights

  • Enforces action-first, step-numbered output style
  • Installable via plugin marketplaces; low adoption friction
  • No visible contributors or releases; maintenance status unclear
  • Repository metadata incomplete (languages/license ambiguous); verify before integration

🔧 Engineering

  • Constrains assistant replies to action-first, numbered steps and explicit next actions to enable quick execution
  • Provides install and invocation examples for multiple coding agents (Claude/Codex), making it easy to adopt

⚠️ Risks

  • Low community activity (no contributors, no releases); long-term maintenance and security updates uncertain
  • Repository metadata is inconsistent: overview shows unknown license while README states MIT; confirm license before production use

👥 For who?

  • Developers/teams using coding LLMs or assistant plugins who want more actionable, concise responses
  • Engineers and reviewers who prefer direct action items and reduced verbosity