skills.

active

Agent skills I use every day

$ cat github.com/arnonsang/skills/README.mdsynced 2026-11-19 10:33 ICT · rebuilt daily

iamickdev skills

Agent skills I use every day, packaged for the skills CLI.
Works with Claude Code, Codex, Cursor and the other agents the CLI supports.

GitHub arnonsang iamickdev.com MIT license


From requirement to product ready

This is a gate-based workflow, not a command chain. Follow only the branches the work needs. A ready artifact permits the next phase; it does not authorize publishing, external communication, or a fleet launch.

flowchart TB
  REQ["Requirement, product idea,<br/>change request, or reported need"]

  subgraph DECIDE["1. Decide what should be built"]
    DR["decision-room<br/>Research facts, expose assumptions,<br/>and record accepted choices"]
    KNOW{"Required knowledge<br/>owned by someone else?"}
    QB["question-brief<br/>Prepare focused stakeholder questions"]
    ANSWERS["Sourced stakeholder answers"]
    UI_DIR{"UI direction or<br/>redesign still uncertain?"}
    DS["design-scout<br/>Research current signals and<br/>product-fit directions"]
    DECISIONS["Accepted decision record"]

    DR --> KNOW
    KNOW -- Yes --> QB --> ANSWERS --> DR
    KNOW -- No --> UI_DIR
    UI_DIR -- Yes --> DS --> DR
    UI_DIR -- No --> DECISIONS
  end

  subgraph DEFINE["2. Define and slice the work"]
    BB["build-brief<br/>Create the governing requirements,<br/>acceptance, and verification seams"]
    BRIEF{"Brief READY?"}
    SM["slice-map<br/>Create dependency-ordered vertical slices,<br/>AFK or HITL ownership, and shared seams"]
    MAP{"Slice map FLEET READY?"}

    BB --> BRIEF
    BRIEF -- Yes --> SM --> MAP
  end

  subgraph BUILD["3. Build"]
    MODE{"Execution mode"}
    DIRECT["Single-agent or team execution"]
    HF["herdr-fleet<br/>Isolated parallel workers,<br/>verification, and integration"]
    CANDIDATE["Integrated product candidate"]

    MODE -- Direct --> DIRECT --> CANDIDATE
    MODE -- "Fleet explicitly requested<br/>and preflight passes" --> HF --> CANDIDATE
  end

  subgraph PROVE["4. Prove readiness"]
    BE["birdseye<br/>Independent outcome and<br/>system-path review"]
    DW["driftwatch<br/>Compare decisions, brief, docs,<br/>code, tests, and observed behavior"]
    UI_WORK{"User interface in scope?"}
    SL["second-look<br/>Remove generic visual habits<br/>and encode project-specific Don'ts"]
    PP["pixel-proof<br/>Prove accessibility, interaction,<br/>responsiveness, and fidelity"]
    UI_NA["UI evidence not required"]
    GATE{"All required evidence passes?<br/>No unresolved critical finding or drift"}
    FIX["Rework the implementation"]
    CAUSE{"Cause unclear, disputed,<br/>or incident-level?"}
    ED["engineering-detective<br/>Route the investigation"]
    BH["bughound<br/>Reproduce, trace, falsify,<br/>fix the cause, and verify"]
    CF["case-file<br/>Preserve a supported bug,<br/>incident, or failed-change record"]
    READY["PRODUCT READY<br/>Accepted behavior is implemented,<br/>verified, aligned, and reviewable"]

    CANDIDATE --> BE --> GATE
    CANDIDATE --> DW --> GATE
    CANDIDATE --> UI_WORK
    UI_WORK -- Yes --> SL --> PP --> GATE
    UI_WORK -- No --> UI_NA --> GATE
    GATE -- Yes --> READY
    GATE -- No --> CAUSE
    CAUSE -- No --> FIX --> CANDIDATE
    CAUSE -- Yes --> ED --> BH --> FIX
    ED -. "Independent fix review" .-> BE
    ED -. "Fixed bug, incident,<br/>or near miss" .-> CF
    CF -. "Record required by the gate" .-> GATE
  end

  subgraph CONTINUITY["Cross-cutting continuity"]
    PAUSE["Any phase pauses, changes owner,<br/>or needs human or management review"]
    CB["continuity-brief<br/>Package verified state for the target"]
    TARGET["Agent, contributor, reviewer,<br/>management, or mixed audience"]

    PAUSE --> CB --> TARGET
  end

  REQ --> DR
  DECISIONS --> BB
  BRIEF -- "No: resolve the named<br/>decision or evidence gap" --> DR
  MAP -- Yes --> MODE
  MAP -- "No: resolve the named<br/>decision, evidence, or dependency" --> DR

  DECISIONS -. checkpoint .-> PAUSE
  BB -. checkpoint .-> PAUSE
  SM -. checkpoint .-> PAUSE
  CANDIDATE -. checkpoint .-> PAUSE
  READY -. handoff .-> PAUSE

  classDef input fill:#f4f0ff,stroke:#6a32de,color:#18131f;
  classDef skill fill:#fff7fb,stroke:#eb147c,color:#18131f;
  classDef gate fill:#fff7e8,stroke:#c47700,color:#18131f;
  classDef ready fill:#ebfff4,stroke:#168353,color:#10251b,stroke-width:2px;
  class REQ,ANSWERS,DECISIONS,CANDIDATE,UI_NA,TARGET input;
  class DR,QB,DS,BB,SM,HF,BE,DW,SL,PP,ED,BH,CF,CB skill;
  class KNOW,UI_DIR,BRIEF,MAP,MODE,UI_WORK,GATE,CAUSE gate;
  class READY ready;

The product-ready gate means the accepted behavior is present, the relevant automated and human checks pass, requirements and implementation do not have unexplained drift, and every critical review finding is resolved or explicitly accepted. UI work also needs second-look and pixel-proof evidence. A substantive bug or incident needs a supported case-file. Use continuity-brief at any checkpoint where another person or agent must safely continue the work.

Skills

Category Skill What it does
Planning decision-room Researches an uncertain idea, asks decisions in dependency order, and routes the accepted result to the right deliverable.
Planning build-brief Writes a decision-complete product and engineering source of truth with traceable requirements and verification seams.
Planning question-brief Prepares focused stakeholder questions when another person owns knowledge required by the work.
Planning slice-map Converts an accepted brief into dependency-ordered AFK and HITL work slices and checks fleet readiness.
Planning continuity-brief Creates a verified, target-aware handoff for a person, agent, reviewer or management audience.
Engineering bughound Diagnoses hard bugs by reproducing, tracing, falsifying and verifying the causal fix.
Engineering birdseye Reviews plans, designs, PRs and code changes through the real system path.
Engineering driftwatch Audits whether requirements, documentation, implementation and observed behavior still agree.
Engineering case-file Writes evidence-based bug, investigation and incident records with optional document output.
Engineering engineering-detective Routes causal diagnosis, independent review and durable engineering records.
Design design-scout Researches current and emerging interface signals before design work.
Design second-look Reviews and repairs interfaces that repeat machine-made visual habits.
Design pixel-proof Verifies UI behavior, accessibility, responsive layout and visual fidelity with evidence.
Orchestration herdr-fleet Runs isolated parallel coding workers through Herdr and verifies every slice before integration.

Each skill’s folder has its own README with what it does, why it exists, and how it was tested.

The planning path is composable: decision-room settles choices, question-brief obtains missing stakeholder knowledge, build-brief creates the governing artifact, slice-map prepares dependency-aware work, and herdr-fleet performs repository-specific parallel execution only after the work is fleet ready.

The engineering path remains separate: driftwatch identifies disagreement, bughound finds and proves a cause, birdseye independently reviews a change, and case-file records supported engineering facts. engineering-detective routes multi-phase investigations.

Install

# every skill in this repo
npx skills add arnonsang/skills

# one skill
npx skills add arnonsang/skills --skill <skill-name>

# list what is inside first
npx skills add arnonsang/skills --list

Adding a skill

skills/
  <category>/
    README.md       # category index, never a SKILL.md
    <skill-name>/
      SKILL.md      # required: frontmatter and instructions
      README.md     # public overview and installation
      scripts/      # optional deterministic helpers
      assets/       # optional output templates
      references/   # optional guidance loaded on demand
      evals/        # optional behavioral prompts
  1. Create skills/<category>/<skill-name>/SKILL.md. The frontmatter name must match the skill folder, and the description says what the skill does and when an agent should use it.
  2. Add README.md next to it, using the same sections as the existing skills.
  3. Keep category folders free of SKILL.md files because a shallower skill shadows nested skills.
  4. Keep it free of private details: no internal hostnames, credentials, customer data or personal paths.
  5. Check the CLI finds it: npx skills add . --list.
  6. Add a row to the Skills table and its category index.

About

I’m Arnon Sang-ngern (@arnonsang, iamickdev), a developer who likes building tools that make daily work easier and more elegant. More of my projects are at iamickdev.com/projects. Reach me at contact@iamickdev.com.

Issues and pull requests are welcome.

License

MIT

continue reading on github →