dangerous-actions

Rosetta CRITICAL MUST skill. MUST activate when action or its consequence is potentially dangerous, potentially irreversible, potentially destructive, or HIGH RISK. MUST activate when consequence MAYBE dangerous even if action itself seems safe. This is enterprise environment — the cost of dangerous activities is EXTREMELY HIGH, recovery may be impossible, and blast radius may affect production, shared environments, or other teams. If there is even a remote chance - load the skill.

Quality

—

Does it follow best practices?

Impact

—

No eval scenarios have been run

Securityby

Passed

No known issues

Quality

Content

72%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is highly actionable with concrete tiers, an exact override mechanism, and explicit approval gates, and it is sensibly self-contained for a safety skill. Its main weaknesses are muddled process numbering that conflates exceptions with steps and some ALL-CAPS redundancy that could be tightened.

Suggestions

Restructure <process> so numbered items are only sequential actions; move Examples and Exceptions into clearly labeled sub-lists so the action sequence is unambiguous.

Reduce ALL-CAPS 'MUST'/'CRITICAL' repetition and deduplicate the blast-radius framing between <process> and <hook> to recover token budget.

Consider moving the Threat model table to a reference file if the skill grows, keeping the SKILL.md overview lean.

Dimension	Reasoning	Score
Conciseness	The body is focused and assumes Claude's competence (no basic-concept explanation), but ALL-CAPS directives and some redundancy between the <process> and <hook> sections, plus an elaborate threat-model table, leave room to tighten.	2 / 3
Actionability	Guidance is concrete and copy-paste ready: specific tiered command examples ('rm -rf /', 'mkfs', 'dd of=/dev/', 'curl \| sh'), an exact override token ('Rosetta-AI-reviewed'), and per-tool field lists (Bash command, Write content, Edit new_string, MCP command/sql/query) with precise detection rules.	3 / 3
Workflow Clarity	The override and hard-deny flows are clear sequences with explicit validation/approval checkpoints and a feedback loop, but the top-level <process> muddles examples and exceptions into the step numbering (steps 5-7 are exceptions, not actions), creating sequence ambiguity in a destructive-operation context.	2 / 3
Progressive Disclosure	No bundle files exist and the skill is a single self-contained safety gate; content is well-organized into signaled sections (Two-tier policy, Threat model, Override mechanism, Hard-deny tier) with no nested references, which is appropriate for in-context loading.	3 / 3
	Total	10 / 12 Passed

Description

57%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description has a strong, explicit activation trigger and a distinct safety niche, but it over-invests in directive "MUST activate" language and never states what the skill actually does. Tightening toward natural user phrasing and adding a concrete capability statement would raise specificity and completeness.

Suggestions

Add a concrete "what this does" clause (e.g., 'Assesses blast radius, requires explicit approval, and proposes safer alternatives for risky actions') so both what and when are answered.

Soften directive 'MUST activate' repetition into natural trigger phrasing users would actually say (e.g., 'Use when an action could be destructive, irreversible, or high-risk').

Include common natural variations of risk terms (e.g., 'production data', 'shared infrastructure', 'cannot be undone') to broaden trigger coverage.

Dimension	Reasoning	Score
Specificity	The description names the domain clearly ("potentially dangerous, potentially irreversible, potentially destructive, or HIGH RISK") but lists no concrete actions the skill performs; it characterizes the risk surface rather than enumerating what the skill does.	2 / 3
Completeness	The "when" is explicit and strong ("MUST activate when action or its consequence is potentially dangerous... If there is even a remote chance - load the skill"), but the "what does this do" is essentially absent — it says "load the skill" without stating what the skill actually does, so it does not clearly answer both.	2 / 3
Trigger Term Quality	It includes relevant risk keywords ("dangerous", "irreversible", "destructive", "HIGH RISK", "blast radius") a user might say, but the phrasing is directive instruction-to-Claude ("MUST activate") rather than natural user trigger language, and common variations are thin.	2 / 3
Distinctiveness Conflict Risk	It carves a clear niche (dangerous/irreversible/destructive enterprise actions with blast-radius concern) with distinct triggers unlikely to be claimed by other skills; the deliberately broad "even a remote chance" is a trigger-calibration choice rather than an overlap with a competing skill.	3 / 3
	Total	9 / 12 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: griddynamics/rosetta
Commit: 1efa337

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.