Static source inference · medium confidence · detected: Sampling
Deprecated protocol patterns detected
Summary
Strong schema coverage with Zod validation and clear naming conventions. Tool names are action-oriented (list_, get_, create_, update_, delete_, generate_, share_, move_). Most parameters include type definitions and descriptions. However, descriptions vary in quality, some are excellent with warnings (get_resource), others are minimal. Output schemas are not explicitly documented. Error handling guidance is limited. The server demonstrates good foundational quality but lacks LLM-optimized descriptions and response documentation needed for A-tier scoring.
Tools (16)
create_folderwriteauthsource verified80/100
Create a new folder.
create_resourcewriteauthsource verified83/100
Create a new password resource.
delete_folderreversibleauthsource verified83/100
Delete a folder (resources inside are NOT deleted, they move to root).
Output schemas not documented. Tool descriptions specify what is returned (e.g., 'List all accessible password resources') but do not document the structure of returned objects, field types, or pagination. LLMs cannot infer response structure and must guess at field names for downstream tool calls.
Minimal descriptions for discovery tools. list_folders, list_users, list_groups have 3-4 word descriptions ('List all X'). No context on when to call them, what data structure is returned, or how to use the results in follow-up calls. These underspecified descriptions reduce LLM selection confidence.
No pagination controls visible in list tools. list_resources, list_folders, list_users, list_groups accept no limit, offset, or cursor parameters. If a Passbolt instance has thousands of resources, LLMs will fetch everything, exhausting context window. No guidance on result limits in descriptions.
Recommendations
Document output schemas for all tools. Add a 'Returns' section to each tool description specifying the structure (e.g., 'Returns an object with fields: id (UUID), name (string), username (string), uri (string), description (string), encrypted_password (string), folder_id (UUID or null).'). This is critical for list_resources, list_folders, list_users, list_groups, get_resource, get_resource_permissions, and get_status.
Add pagination support to list_resources, list_folders, list_users, list_groups. Include optional 'limit' (default 20, max 100) and 'offset' or 'cursor' parameters. Update descriptions to state: 'Returns up to [limit] items. Use offset/cursor for pagination to avoid context exhaustion.' This prevents LLMs from fetching thousands of records.
Enhance descriptions for discovery tools. Change 'List all accessible folders' to 'List all accessible folders. Call this first to see available folder structure before creating or moving resources. Returns folder name, ID, and hierarchy.' Similar clarity needed for list_users and list_groups.
Add error classification and recovery guidance in tool descriptions. For delete_resource, add: 'May fail with PermissionDenied (check permissions via get_resource_permissions) or ResourceNotFound. Deletion is irreversible.' For share_resource, add: 'Fails if resource_id or user_id is invalid or if caller lacks Owner permission.'
Document permission type values in share_resource. Change 'type (1=Read, 7=Update, 15=Owner)' description to include: '1 = Read-only (view password), 7 = Update (modify and view), 15 = Owner (full control and re-share).' This removes ambiguity for LLMs.
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Sampling (deprecated) - integrate directly with the LLM provider API
Score history
Overall score trend
↑ 49 points across a rubric change (v1 → v2)
60/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
C
60
<=2025-11-25
v2
2026-03-09
F
11
-
v1
80/100
List current permissions (who has access) for a resource.
get_statusread onlyauthsource verified77/100
Check Passbolt server status and API version.
list_foldersread onlyauthsource verified78/100
List all accessible folders.
list_groupsread onlyauthsource verified77/100
List all groups in the Passbolt instance.
list_resourcesread onlyauthsource verified83/100
List all accessible password resources, optionally filtered by folder.
list_usersread onlyauthsource verified77/100
List all users in the Passbolt instance (needed for sharing).
move_resourcereversibleauthsource verified82/100
Move a resource to a different folder (or to root).
share_resourcewriteauthsource verified88/100
Share a resource with users or groups. Sets permissions and re-encrypts the secret.
update_folderwriteauthsource verified82/100
Rename a folder.
update_resourcewriteauthsource verified82/100
Update an existing password resource (metadata and/or password).
Error handling lacks recovery guidance. The CallToolRequestSchema handler catches errors and returns isError: true with a message, but does not categorize errors (retryable vs user-fixable vs fatal) or guide the LLM on next steps. E.g., if delete_resource fails due to permission denial, the LLM has no instruction to check permissions or call list_users.
Destructive operations (delete_resource, delete_folder) lack confirmation or dry-run patterns. No tool supports a 'confirm' step or preview. If an LLM misinterprets a user's intent, deletion is irreversible without recovery tooling.
Incomplete parameter descriptions for constrained inputs. share_resource's 'permissions' parameter documents the array structure with enum values for 'type' (1, 7, 15) but does not explain what those integers mean in user-friendly terms (Read, Update, Owner). An LLM must reverse-engineer the mapping or guess.
No documented chaining IDs in responses. The tools return resources, folders, and users, but documentation does not specify which fields (e.g., resource_id, folder_id, user_id) are guaranteed in responses. Downstream tools like share_resource expect resource_id, but if get_resource returns 'id' instead, the LLM must infer the mapping.
Tool annotations missing. No tools declare readOnlyHint, destructiveHint, or idempotentHint. Without these hints, LLMs cannot automatically determine safe retry logic or whether an operation modifies state. E.g., list_resources should hint readOnly=true; delete_resource should hint destructive=true.
Add tool annotations to the Server initialization. When registering tools via setRequestHandler, decorate each with readOnlyHint, destructiveHint, or idempotentHint. Example for list_resources: annotate with readOnly: true. Example for delete_resource: annotate with destructive: true. This enables better LLM planning and retry logic.
Add confirmation support for destructive operations. Implement a pattern where delete_resource and delete_folder support an optional 'confirm' parameter. If confirm is false (default), return a preview message 'Deleting resource XYZ, call again with confirm: true to proceed.' This prevents accidental deletions.
Document natural identifiers. Extend tools to accept both UUIDs and human-readable names where practical. E.g., move_resource could accept folder_id as UUID or folder_name as string, resolving names to IDs internally. Add to description: 'folder_id may be a UUID or folder name.'
Enhance generate_password description. Current text is minimal. Change to: 'Generate a secure random password for new resources. Customize length (8-128 chars, default 20) and character composition. Useful before calling create_resource or update_resource.'
Add dependency hints. In create_resource, add: 'If you only have a folder name, call list_folders first to get the folder_id.' Similar hints for share_resource: 'If you only have a username, call list_users first to get the user ID.'
Document pagination in list tool responses. E.g., list_resources should state: 'Returns {total: number, count: number, resources: [{id, name, username, uri, ...}]}. Use limit and offset for large result sets.' This helps LLMs understand how to iterate.
Add validation error guidance. In get_resource, add to description: 'Fails if id is not a valid UUID or if the resource does not exist or is not accessible.' Similar specificity for all parameter-dependent tools.
Clarify include_secret behavior. In get_resource, expand: 'If include_secret is true (default), the decrypted password is returned in plaintext in the response, visible to the LLM and any logs. Set include_secret: false to omit the password. WARNING: The decrypted password will be visible in the AI context.'
Document move_resource null handling. Clarify that folder_id: null moves the resource to root (top level), not to a delete/archive state. Current description does not specify the null case.
Add response field naming documentation. Ensure all tools document which fields are guaranteed (e.g., 'id' vs 'resource_id'). Standardize on a naming scheme (e.g., always use 'id', 'folder_id', 'user_id') so downstream tools can rely on consistent field names.