H4tcher Content Author

> Last updated: 2026-05-29

Content Author

Author or modify a hatch3r canonical content artifact.

§0.1: Ambiguity Gate (P8 B1)

Per CONSTITUTION §2 P8 B1: every framework-dev workflow mutating canonical artifacts detects and resolves ambiguity via agents/shared/user-question-protocol.md BEFORE executing — default behavior, not exception-driven. Apply the protocol when any hold: (a) ambiguous scope (which artifact type — agent vs skill vs command per the Decision #13 discriminator — or new-vs-edit), (b) multiple valid interpretations with materially different cost/scope/risk, (c) irreversible action (public artifact id rename, frontmatter-field drop), or (d) missing acceptance criteria (no testable definition of done). Use the platform-native question tool; one question per turn; 2-4 numbered options with one-line trade-offs; declare the default-if-no-response. Do NOT proceed to Step 1 until resolved or the default is confirmed.

Step 1: Determine Artifact Type

Identify the content type from user input:

• Agent: agents/hatch3r-{name}.md
• Skill: skills/hatch3r-{name}/SKILL.md
• Rule: rules/hatch3r-{name}.md + rules/hatch3r-{name}.mdc (both required)
• Command: commands/hatch3r-{name}.md
• Hook: hooks/hatch3r-{name}.md

Step 2: Study Existing Patterns

1. Read agents/shared/quality-charter.md — all artifacts must embody these standards
2. Read 2-3 existing artifacts of the same type for frontmatter schema, structure, tone
3. Check agents/shared/external-knowledge.md for tooling hierarchy (if it exists)

Step 3: Reputable-Source Reconnaissance (Decision #14)

Required for new agents, skills, and rules. Skip only for trivial edits (typo, frontmatter-only, single-line clarification).

1. Web-search ≥2 independent reputable sources covering the topic the new artifact will address. Trust tier ordering per agents/shared/rigor-contract.md: official vendor docs > peer-reviewed methodology > named-maintainer agent/skill libraries (e.g., Anthropic Claude Code agent registry, established cookbook patterns) > vendor notes > independent analyses > blog posts. Recency window ≤12 months.
2. Synthesise patterns — what structural choices, decision criteria, and verification steps recur across sources. Do NOT copy verbatim; encode the synthesis in the new artifact's voice.
3. Record sources in a ## References section at the bottom of the new artifact: URL, access date (YYYY-MM-DD), trust tier, one-line attribution of what you took from each.
4. If <2 reputable sources exist for the topic, halt and surface that finding — the artifact may be premature.

Step 4: Author

1. Write YAML frontmatter:

``yaml --- id: hatch3r-{name} type: agent|skill|rule|command|hook description: One-line description tags: [relevant, tags] quality_charter: agents/shared/quality-charter.md --- ``

1. For agents: include role definition, key files, key specs sections
2. For skills: follow Quick Start + Step pattern with numbered workflow steps
3. For rules: produce both .md and .mdc variants with matching content
4. For commands: include workflow steps, sub-agent delegation, quality gates

Step 5: Validate

1. Pillar alignment: Document which pillar(s) this artifact serves (P1-P8)
2. Duplication check: Search existing artifacts for overlapping scope:

`` grep -r "similar-keyword" agents/ skills/ rules/ commands/ hooks/ `` If overlap found: propose merge or boundary clarification instead of creating duplicate

1. D05 universal checklist (from governance/audit/domains/D05-prompt-engineering.md):

• Would an LLM execute this on first attempt without clarification?
• Are instructions unambiguous and sequenced logically?
• Is expected output format explicitly defined?
• Are scope boundaries clear?
• Is the artifact sized to the limit for its type? There is no universal <150 cap. Pipeline agents (research/implement/review/final) target ≲150 lines per D05 §5.1; every other class follows its CONSTITUTION §2 P5 file-class limit (e.g., rules/*.md critical/high ≤250, normal/low ≤120). Orchestrators, board, and spec artifacts legitimately run longer — measure against the class limit, not 150.

1. Anti-slop scan: Verify zero banned phrases per wordlist in CLAUDE.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hatch3r
• Source: hatch3r/hatch3r
• License: MIT
• Homepage: https://docs.hatch3r.com

Install and usage instructions live in the source repository linked above.