values-firewall

check_action

Kerne-gaten (to-lags evaluering): Lag A (deterministisk hard-rule/outward-match, ingen LLM) + Lag B (Sonnet vurderer mod charterets NL-principper via LiteLLM :4000, citerer principle_id + reasoning — INGEN egen alvorligheds-dom) → most-restrictive-wins aggregeret, auditerbar receipt: {action, deterministic_flags[], values_assessment[], values_assessment_unavailable, unrecognized_principle_citations, principles_evaluated, recommendation, requires_human, charter_name, scope_note}.

recommendation ∈ {allow, flag, gate_to_human} — ALDRIG approved/denied/true/false. ADVISORY: anbefaler, blokerer/eksekverer/autoriserer ALDRIG (never-decide-for-you). FAIL-SAFE: Lag B utilgængelig -> Lag A kører stadig, values_assessment_unavailable=true, recommendation løftet til MINDST 'flag' (aldrig stille 'allow'). Read-only — skriver INTET, ingen side-effekter.

PUBLIC_DEMO-mode: vurderet mod det OFFENTLIGE Art-14-demo-charter (ikke din org — se scope_note). Rate-limitet (default 5/min + 30/time, ÉN global demo-bucket). Over grænsen returneres {error:"rate_limited", retry_after_s, scope_note} — ALDRIG et recommendation-formet svar. Bogholderi-fejl i rate-limiteren -> deny (fail-safe).

RED-TEAM-FIX (minimal-indgreb): tomt/whitespace-only action valideres FØR rate-limiteren forbruger en kvote-token — en ugyldig-input-loop kan ellers stille dræne demoens kvote uden at nogensinde levere en receipt (rå ValueError fra firewall.check_action's fail-loud-guard ville ellers både forbruge kvote OG propagere ubehandlet til transport-laget, hvilket bryder "check_action leverer ALTID en receipt/struktureret fejl"-disciplinen). Uden for PUBLIC_DEMO uændret: fw.check_action's egen ValueError-fail-loud gælder stadig (bindende, testet i test_values_firewall.py).

Input Schema

TableJSON Schema

Name	Required	Description	Default
`action`	Yes
`context`	No

Output Schema

TableJSON Schema

Name	Required	Description	Default
No arguments

TDQS

A3.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description fully discloses key behavioral traits: read-only with no side effects, advisory-only (never blocks/executes), most-restrictive-wins aggregation, fail-safe when Layer B is unavailable (recommendation lifted to at least 'flag'), rate limiting behavior, and error handling (rate_limited response). It also details the red-team-fix for empty action validation, which is highly transparent about edge cases.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness2/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long and dense, with multiple sections including a very specific RED-TEAM-FIX that is tangential to general tool usage. While it is structured in paragraphs, it lacks bullet points or clear separation of core vs. edge-case information, making it harder to parse quickly. Every sentence adds some behavioral detail, but the overall size is excessive for a tool selection scenario.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The description is highly comprehensive for a tool with an output schema: it explains the returned receipt fields, recommendation values, rate-limiting errors, and fail-safe modes. However, it leaves a gap in parameter semantics (especially 'context') and does not fully describe what constitutes a valid 'action' string, which is a notable omission for practical use.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The description provides minimal parameter meaning: it discusses 'action' only in the context of empty/whitespace validation, and does not explain what 'action' should contain or what 'context' is. Input schema coverage is 0%, and the description does not compensate by describing parameter semantics beyond that single edge case.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's core function: a two-layer evaluation gate (deterministic + LLM) that produces a recommendation (allow/flag/gate_to_human) for an action. It distinguishes itself from the sibling check_action_deterministic by explicitly describing the dual-layer architecture and the advisory nature.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description implies when to use the tool (e.g., when a full two-layer evaluation is needed, or when operating in PUBLIC_DEMO mode) and describes fail-safe behavior, but it does not explicitly state when to prefer this over check_action_deterministic or list_charter. No explicit exclusion guidance is provided beyond mentioning different modes.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

All values-firewall tools Inspect

Try in Browser

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

TDQS

A4.3/5.0

Disambiguation5/5

Each tool has a distinct role: full values assessment, deterministic assessment without LLM, and charter listing. The overlap between the two assessment tools is clearly bounded by the explicit layer difference (Lag A vs Lag A+B).

Naming Consistency5/5

All tools use verb_noun underscore naming (check_action, check_action_deterministic, list_charter). The deterministic variant logically extends check_action, maintaining a predictable pattern.

Tool Count5/5

Three tools is well-scoped for a read-only advisory firewall. Each tool serves a necessary function without unnecessary bloat.

Completeness5/5

The domain is fully covered: inspect the governing charter, run a full assessment, and run a fast deterministic assessment. No obvious missing operations for the stated purpose.

Resources

Need Help?