Use of Claude-specific features hinders portability

Several skills reference Claude-specific tool interfaces, workflow runtimes, hooks, or credentials. These assumptions raise portability concerns for other agent environments. This checklist identifies dependencies to adapt or document; it does not establish that every listed skill fails elsewhere.

When checking off an item, please link the relevant PR in a comment, or reference this issue from that PR.

Checklist

Workflow execution and lifecycle

• audit-context-building — The main codebase-analysis path invokes a dynamic workflow whose implementation calls the host-provided agent() function. Routing · Workflow implementation
• spec-to-code-compliance — The main compliance-analysis path invokes a dynamic workflow using host-provided agent() and parallel() functions. Routing · Workflow implementation
• c-review — Requires the Workflow tool and prohibits substituting manually spawned agents; also names AskUserQuestion and the Claude models haiku, sonnet, and opus, with an additional inherit option. Workflow requirement · Questions and model choices
• code-improver — Requires Workflow and stops if it is unavailable; names TaskOutput for awaiting completion and TaskStop for stopping a run. Workflow requirement · Waiting · Stopping
• pr-improver — Requires Workflow and stops if it is unavailable; names TaskOutput for awaiting completion and TaskStop for stopping a run. Workflow requirement · Waiting · Stopping
• skill-improver — Requires Workflow, names TaskOutput and TaskStop, and uses plugin-dev:skill-reviewer from the claude-plugins-official marketplace. Workflow, waiting, and reviewer · Stopping · Reviewer identifier
• semgrep-rule-variant-creator — Uses a dynamic workflow and an explicit Workflow(...) resume call; recovery instructions inspect Claude session-specific workflow storage. It also documents path lookup for multiple agent environments. Workflow and recovery · Path lookup
• insecure-defaults (workflow-only plugin) — Its audit command requires Workflow and uses CLAUDE_PLUGIN_ROOT to locate resources; workflow phases specify the Claude models sonnet and opus. Command · Model choices

Agent dispatch and interaction

• zeroize-audit — Phase instructions explicitly spawn agents through Task with named subagent_type values; the orchestrator also names AskUserQuestion for verification failures. Task dispatch · Question tool
• rust-review — Uses Agent(subagent_type=...), SendMessage, AskUserQuestion, and task-management tools; worker model choices are haiku, sonnet, and opus. Agent calls and messaging · Questions and model choices · Tool declarations
• cosmos-vulnerability-scanner — Explicitly requests synchronous workers through the Agent tool with a bundled discovery prompt. Source
• github-triage — Names AskUserQuestion for repository selection and approvals, and Agent for parallel PR reviews. Repository selection · Approvals · Review agents
• codeql — Explicitly names AskUserQuestion for database and workflow selection; frontmatter also lists TaskCreate, TaskList, TaskUpdate, and TaskGet. Questions · Tool declarations
• semgrep — Uses AskUserQuestion for approval and explicitly requires TaskCreate tasks with blockedBy dependencies. Task dependencies · Approval tool
• testing-handbook-generator — Supporting discovery instructions explicitly use TodoWrite for progress tracking. Source

Plugin hooks and environment integration

• fp-check — The plugin supplies prompt-type Stop and SubagentStop hooks for output-completeness checks. Stop hook · SubagentStop hook
• gh-cli — Plugin SessionStart setup relies on CLAUDE_ENV_FILE to install PATH shims; interception hooks emit Claude PreToolUse permission-response fields. Session setup · Interception response
• modern-python — Plugin SessionStart setup relies on CLAUDE_ENV_FILE to install PATH shims. Source
• slicing-code-context — Its script-path fallback uses CLAUDE_SKILL_DIR; execution instructions stop on failure rather than substituting manually selected source. Script path and failure behavior

Optional workflows and commands

• post-patch-validation — Provides an explicitly labeled Claude dynamic-workflow interface using Workflow(...), alongside its Python runner. Optional workflow interface
• variant-analysis — Offers a dynamic workflow using runtime-provided agents; its resource lookup checks CLAUDE_PLUGIN_ROOT and Claude plugin-cache paths before broader filesystem searches. The skill also documents direct execution of its analysis steps. Optional workflow · Resource lookup · Agent invocation
• git-cleanup (command) — Names Workflow, CLAUDE_PLUGIN_ROOT, and AskUserQuestion. It documents an inline analysis fallback when the workflow is unavailable or its path cannot be resolved. Workflow and path setup · Inline fallback · Question tool

Container configuration

• devcontainer-setup — Generated container configuration forwards CLAUDE_CODE_OAUTH_TOKEN and ANTHROPIC_API_KEY; the bundled post-install helper invokes Claude onboarding. Credentials · Onboarding helper

Scope and metadata review

These items concern metadata or a skill dedicated to a Claude integration; their presence alone does not establish a functional portability problem.

• second-opinion — Lists AskUserQuestion in allowed-tools frontmatter; the body uses generic instructions for collecting missing choices. This is a metadata-only review item, not an established runtime dependency. Source
• chrome-mcp-troubleshooting — Specifically supports mcp__claude-in-chrome__*, the Claude browser extension, and its native messaging integration. Tool scope · Native messaging integration

---

Source links are pinned to public-repository commit 0f44fe01dcb00bd6929b4ba82a257d305de04554, inspected on October 14, 2026.