translate-copilot-to-claude

name

translate-copilot-to-claude

description

Translate one or more GitHub Copilot native files (.github/copilot-instructions.md, .github/instructions/*.instructions.md, .github/agents/*.agent.md, .github/prompts/*.prompt.md, .github/skills/<name>/SKILL.md) into the native Claude runtime. Classify each section into agent/rule/skill/hook surfaces, diff against existing .claude/ state, produce a translation plan for user approval, and then apply the plan.

Translate Copilot to Claude

Deterministic translation workflow that ports GitHub Copilot native files into the .claude/ runtime without duplicating content, overwriting user edits, or breaking existing agents.

When to Use This Skill

Use this skill when:

• One or more .github/ native files must be mirrored into .claude/.
• An agent persona needs to be split across Claude's four runtime surfaces.
• A batch of instructions files must be re-expressed as path-scoped Claude rules.
• A Copilot prompt or skill must be reshaped into a Claude skill.
• The user wants to know what is missing or out-of-date on the Claude side before making changes.

Inputs

Required:

• One or more source paths under .github/. Accepted types:
  • .github/copilot-instructions.md
  • .github/instructions/<name>.instructions.md
  • .github/agents/<name>.agent.md
  • .github/prompts/<name>.prompt.md
  • .github/skills/<name>/SKILL.md

Optional:

• mode=plan-only — do not write anything; emit the translation plan and stop.
• mode=apply — apply the plan after it is produced and approved in the same turn.
• target-scope=<agent|rule|skill|hook|all> — restrict surfaces considered for this run.
• timestamp-override=<ISO-8601> — override the auto-generated evidence folder timestamp.

Default mode is plan-only. Apply requires explicit confirmation either via mode=apply or a plain-text approval in the same turn.

Phase 1 — Intake and Section Extraction

For each input file:

1. Read the full file with Read.
2. Parse YAML frontmatter (name, description, tools, applyTo, agent, handoffs, model, other keys).
3. Split the body into top-level sections delimited by #, ##, ### headings.
4. Classify each section as declarative, procedural, enforceable, identity, or handoff per the taxonomy in Phase 2.
5. Record each section as {source_path, heading, classification, raw_content} in an in-memory list.

Never modify or delete source files. Translation is a copy-forward operation.

Phase 2 — Classification Taxonomy

Apply these decision rules in order. Use the first rule that matches.

2.1 Enforceable (hard gate / hard stop) -> .claude/hooks/

A section is enforceable when it contains:

• phrases such as "hard gate", "hard stop", "must not", "blocked", "forbidden", "zero-regression", "non-negotiable",
• command or path blacklists that should be rejected regardless of model intent,
• per-batch counting or budget limits,
• toolchain-execution verification ("must run", "no unverified work"),
• test-purity forbiddens (tempfile, network, subprocess in tests).

Target:

• A new or updated script under .claude/hooks/<name>.ps1.
• Registration in .claude/settings.json under hooks.PreToolUse or hooks.SubagentStop with a matcher that targets the affected tools (Bash, Write|Edit, specific agent names).

Hooks must be read-only validation gates unless the user explicitly authorizes state mutation. Prefer exit code 0 with a {"decision":"block","reason":"..."} JSON response over a non-zero exit when feedback to the agent is required.

2.2 Declarative standards (always-on for matching files) -> .claude/rules/

A section is declarative when it contains:

• coding conventions (naming, typing, imports, error handling),
• test structure requirements,
• toolchain ordering statements,
• file-size limits,
• language-specific style rules.

Target:

• An existing rule file under .claude/rules/ that already covers the same path scope, merged into place.

• A new rule file under .claude/rules/<language-or-topic>.md with frontmatter:

  ---
  paths:
    - "<glob>"
  description: "<one line>"
  ---
  

• If the source frontmatter used applyTo: "<glob>", convert to paths: [<glob>].

2.3 Reusable procedures -> .claude/skills/<name>/SKILL.md

A section is procedural when it contains:

• multi-step workflows ("Phase A", "Phase B", "Workflow", numbered steps),
• reporting templates,
• routing decision tables,
• delegation contracts.

Target:

• An existing skill under .claude/skills/ when one already covers the same concept (check .claude/skills/<name>/SKILL.md by name and description).
• A new skill at .claude/skills/<name>/SKILL.md when no coverage exists. Use the make-skill-template skill as the scaffold if needed. Frontmatter must include name and description.

Copilot .github/prompts/*.prompt.md and .github/skills/<name>/SKILL.md default to this target. Preserve the original name when it does not collide with an existing Claude skill.

2.4 Identity and preload manifest -> .claude/agents/<name>.md

A section belongs on the persona when it contains:

• agent name, description, role summary,
• tool allowlist,
• model selection,
• skill preloads,
• memory scope,
• a short workflow index pointing to preloaded skills,
• stop conditions.

Target:

• An existing agent file under .claude/agents/<name>.md. Update only the preload list, tools, description, and stop conditions. Keep the body thin.
• A new agent file when the Copilot persona has no Claude counterpart. Do not duplicate rule, skill, or hook content into the persona body.

2.5 Handoffs and delegation -> reference existing skills

Sections describing handoffs to other agents ("delegate to atomic_planner", "handoffs:" frontmatter) should not be duplicated. Reference the canonical skill:

• plan authoring -> atomic-plan-contract
• remediation -> remediation-handoff-atomic-planner
• promotion lifecycle -> feature-promotion-lifecycle
• PR context -> pr-context-artifacts
• evidence storage -> evidence-and-timestamp-conventions

2.6 Repo-wide tone / policy precedence -> CLAUDE.md

Sections that state cross-cutting repository defaults (tone, policy reading order, architectural overview) belong in CLAUDE.md and should be cross-referenced from .claude/rules/ rather than restated.

2.7 Out-of-scope content

The following content is not translated:

• GitHub Actions workflow YAML under .github/workflows/.
• Codex or other non-Copilot customizations under .github/codex/ unless explicitly requested.
• Copilot chatmodes unless the user specifies a target Claude surface.

Phase 3 — Target Resolution

For each classified section, compute a concrete target path:

1. Hook: .claude/hooks/<verb-noun>.ps1 where the verb is check|enforce|validate and the noun is the concern (for example, check-python-test-purity.ps1, enforce-python-batch-budget.ps1).
2. Rule: .claude/rules/<topic>.md where topic is the language (python.md, csharp.md, powershell.md, typescript.md) or cross-cutting concern (tonality.md, general-code-change.md, general-unit-test.md).
3. Skill: .claude/skills/<kebab-case-name>/SKILL.md. Prefer verb-first names for action skills (review-feature, implement-python, invoke-python-engineer) and topic names for contract skills (python-qa-gate, atomic-plan-contract).
4. Agent: .claude/agents/<kebab-case-name>.md. Normalize underscores to hyphens (python_typed_engineer -> python-typed-engineer).
5. CLAUDE.md: single repo-wide file at the workspace root.
6. settings.json: .claude/settings.json sections hooks and permissions.

Record target paths in the plan. Do not write anything yet.

Phase 4 — Existing State Diff

For every target path:

1. Read the existing file when it exists.
2. Compute a per-section delta:
   • add — the section does not exist in the target.
   • replace — the section exists but the source content supersedes it (explicit user intent).
   • merge — the section exists and source content must be appended or integrated without removing existing text.
   • skip — the section is already present verbatim or has a newer canonical form in the target.
   • conflict — the source and target disagree on substantive content; requires user decision.
3. For .claude/settings.json, compute add-only updates to hooks.* arrays and permissions.allow. Never remove existing permissions without explicit instruction.
4. For agent personas already present, prefer merge over replace to preserve hand-authored guidance.

Flag every conflict row in the plan. Conflicts stop Phase 6 for that row until resolved.

Phase 5 — Translation Plan Artifact

Write a translation plan to artifacts/translation/<timestamp>/plan.md. The timestamp is ISO-8601 UTC, for example 2026-04-18T14-30-00Z, per evidence-and-timestamp-conventions.

Plan structure:

# Translation Plan: <source file basename(s)>

Generated: <timestamp>
Mode: <plan-only | apply>

## Inputs
- <source path 1>
- <source path 2>

## Mapping Table
| Source Section | Classification | Target Path | Action |
|---|---|---|---|
| <source#section-anchor> | rule | .claude/rules/python.md#pytest-rules | merge |
| <source#section-anchor> | hook | .claude/hooks/check-python-test-purity.ps1 | add |
| <source#section-anchor> | skill | .claude/skills/python-qa-gate/SKILL.md | add |
| <source#section-anchor> | agent | .claude/agents/python-typed-engineer.md | merge |
| <source#section-anchor> | settings | .claude/settings.json | add |

## Conflicts (require user decision)
<one row per conflict, or "none">

## New Files
<list of new file paths>

## Updated Files
<list of modified file >


 Hooks added: 
 Permissions added: 
 Matchers extended: 


 artifacts/translation//plan.md
 artifacts/translation//diff.md (populated after apply)

Always produce the plan artifact, even in mode=apply runs, for auditability.

Phase 6 — Apply (only after explicit approval)

Only execute when one of the following holds:

• The user invoked the skill with mode=apply.
• The user sent a plain-text approval ("proceed", "apply", "execute the plan") in the same turn after the plan was shown.

Apply order (to minimize breakage):

1. Rules — add or merge rule files under .claude/rules/.
2. Skills — add or merge skill files under .claude/skills/<name>/SKILL.md.
3. Hooks — add hook scripts under .claude/hooks/ and parse-check them with pwsh -NoProfile -Command "[System.Management.Automation.Language.Parser]::ParseFile(..., [ref]$null, [ref]$errors)".
4. Agent personas — update .claude/agents/<name>.md frontmatter (tools, skills, memory) and trim the body to identity + workflow index.
5. settings.json — append hook registrations and Skill(...) permissions. Never rewrite the file; use targeted edits.
6. CLAUDE.md — append cross-references only; never overwrite existing sections.
7. Evidence — write artifacts/translation/<timestamp>/diff.md summarizing what was actually changed, and store a copy of every new or modified file path under artifacts/translation/<timestamp>/snapshots/.

After apply, run a verification sweep:

• Parse-check every new or changed .ps1 under .claude/hooks/.
• JSON-parse .claude/settings.json to confirm validity.
• Confirm every new skill appears in the skills catalog by reading back the first five lines of each target file.
• Report the final mapping table with done / skipped / conflict-unresolved statuses.

Phase 7 — Reporting

Every completion response must include:

1. Inputs — list of source files.
2. Mapping summary — counts by classification (rule, skill, hook, agent, settings).
3. Action summary — counts by action (add, replace, merge, skip, conflict).
4. Files changed — explicit list of created and modified paths.
5. Settings delta — hooks and permissions added.
6. Evidence paths — artifact locations.
7. Conflicts — unresolved conflict rows, or "none".

Guarantees and Prohibitions

• Idempotency: running the skill twice on the same input produces no new changes beyond the regenerated plan artifact.
• No source deletion: .github/ files are never modified or removed.
• No silent overwrites: existing .claude/ content is merged by default. Replace requires either user instruction or a conflict row resolved explicitly.
• No hook-based state mutation: hooks generated by this skill are validation-only unless the user authorizes otherwise.
• No settings.json rewrites: only targeted edits that append permissions or matchers.
• No translation of out-of-scope surfaces: .github/workflows/, .github/codex/, and unrecognized file types are skipped and listed under "Skipped inputs" in the plan.

Classification Quick Reference

Copilot source pattern	Default Claude target	Notes
.github/copilot-instructions.md	CLAUDE.md + .claude/rules/tonality.md	Tone and precedence only.
.github/instructions/*.instructions.md (applyTo)	.claude/rules/<topic>.md (paths)	Direct frontmatter rewrite.
.github/agents/*.agent.md identity frontmatter	.claude/agents/<name>.md frontmatter	Convert tools: to Claude tool names.
.github/agents/*.agent.md "Absolute guardrails / hard gate"	.claude/hooks/*.ps1 + settings.json	Hooks enforce; rules merely document.
.github/agents/*.agent.md "Workflow / Phase X"	.claude/skills/<workflow>/SKILL.md	Split per phase when phases are independent.
.github/agents/*.agent.md "Design rules / Testing rules"	.claude/rules/<language>.md	Merge under Design Rules / Pytest Rules subheadings.
.github/agents/*.agent.md "Handoffs"	Reference existing routing skills	Do not duplicate.
.github/agents/*.agent.md "Reporting requirements"	.claude/skills/<name>-qa-gate/SKILL.md or the QA gate skill	Procedure, not identity.
.github/prompts/*.prompt.md	.claude/skills/<action-name>/SKILL.md	One skill per prompt.
.github/skills/<name>/SKILL.md	.claude/skills/<name>/SKILL.md	Near-1:1; reconcile frontmatter schema.

Invocation Examples

• Translate a single agent persona, plan only: translate-copilot-to-claude .github/agents/python-typed-engineer.agent.md
• Translate a set of instruction files and apply: translate-copilot-to-claude .github/instructions/python-code-change.instructions.md .github/instructions/python-unit-test.instructions.md mode=apply
• Translate an entire agent bundle (persona + its referenced instructions): translate-copilot-to-claude .github/agents/csharp-typed-engineer.agent.md .github/instructions/csharp-code-change.instructions.md .github/instructions/csharp-unit-test.instructions.md target-scope=all