Loading ToolBench

Reading the current corpus…

PiLink MCP server · B · ToolBench

PiLink

OAuth-protected MCP bridge with single-agent, public-chat orchestration, optional VS Code, and CLI model endpoint workflows

2026-07-28+STREAMABLE_HTTP4 tools

75/100

Definition Quality75

Protocol Readiness75

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · detected: Dynamic Client Registration

Deprecated protocol patterns detected

Summary

PiLink implements a coordination-focused agent toolset with 4 tools serving internal agent-to-agent communication and task management. Tool definitions include schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention well. Descriptions exist but lack depth in explaining WHEN to use tools and what downstream tools they enable. Schemas are present with type information but several parameters lack detailed format/constraint descriptions. Error handling is not visible in provided source excerpts. The toolset is self-contained with no integration to external systems, limiting composition complexity but also limiting real-world utility signal.

Tools (4)

coordination_chat_postwriteauthsource verified73/100

Post a message to this project's private agent coordination channel as the current supervised agent.

coordination_chat_readread onlyauthsource verified75/100

Read recent messages from this project's private agent coordination channel using a cursor.

coordination_task_listread onlyauthsource verified77/100

List only coordination tasks assigned to this exact supervised agent.

coordination_task_updatewriteauthsource verified75/100

Update a task assigned to this exact supervised agent using its expected revision.

Issues (7)

HIGHresponse-shaper ↗

Output schemas undocumented. Tools define inputs clearly but responses are not documented in provided source. LLM cannot plan downstream tool calls without knowing what fields each tool returns (e.g., does coordination_chat_post return message_id? timestamp? coordination_task_list return total_count or next_cursor?).

coordination_chat_readcoordination_chat_postcoordination_task_listcoordination_task_update

HIGHconstrained-input ↗

Status enum inconsistency. coordination_task_list accepts 7 statuses (open, working, input_required, blocked, completed, failed, cancelled) but coordination_task_update only permits 4 (working, blocked, completed, failed). LLM may attempt invalid status transitions or filter by statuses that cannot be set (open, input_required, cancelled). Document the state machine.

coordination_task_listcoordination_task_update

HIGHrecovery-guide ↗

Missing error recovery guidance. No tool descriptions explain failure modes or how an agent should recover. E.g., what happens if task_id does not exist? If expected_revision is stale? If status transition is invalid? LLM gets no actionable next step from errors.

Recommendations

Document output schema for each tool. Create a response example or JSON Schema showing: coordination_chat_read returns [{id, author, timestamp, text, ...}], coordination_chat_post returns {id, created_at, status}, coordination_task_list returns {tasks: [{id, status, ...}], total, next_cursor}, coordination_task_update returns {id, status, updated_at, revision}.
Clarify status state machine. Document valid transitions: open → working → (blocked | completed | failed); open → input_required; input_required → (working | cancelled). Explain which statuses are set by update_task and which are system-generated.
Add error recovery hints to all descriptions. E.g., 'If status transition fails (e.g. already completed), call coordination_task_list to verify current status and retry.' 'If expected_revision is stale, fetch the task via coordination_task_list, read its revision field, and retry update.'
Expand parameter descriptions with format examples. E.g., 'message text (UTF-8, supports markdown, max 16384 bytes; newlines and mentions via @agent-name are supported)'. 'artifact data (JSON object or plaintext string, max 32768 bytes; commonly used for task output like test results or logs)'.
Add a section to each tool description explaining 'When to use this tool' and 'What to do next'. E.g., coordination_task_list: 'Use this to discover tasks assigned to you. After fetching task details, call coordination_task_update to report progress or completion.'
Document cursor pagination semantics for coordination_chat_read. Clarify: 'after parameter is the zero-based message index; use the index of the last message retrieved to fetch the next batch. Response includes a has_more flag or next_cursor to indicate if more messages exist.'

Score history

Overall score trend

First recorded score · v2 rubric

75/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	B	75	2026-07-28+	v2