Local evidence recorder and verification layer for AI coding agents.
Critical gaps across all dimensions. 24 tools with minimal descriptions (most 20-50 chars, well below 194-char baseline). Input schemas present but severely underdocumented: parameters lack type constraints, enums, ranges, and validation guidance. No output schemas documented. Error handling absent. Many tools combine multiple concerns (e.g., create_run + record_event + stop_run form a workflow, not atomic operations). Naming inconsistent: 'record_command' vs 'record_event' vs 'snapshot_files' lack clear verb-noun patterns. Archive tools (sidecar, orchestration) are inferred from file paths, not visible in active registration. No security annotations, permission gates, or audit trail guidance.
Append an event to a run via the sidecar HTTP API
Apply an automatic amendment to a policy
Approve a pending approval via the sidecar HTTP API
Build a policy from a template with task details and worker scopes
Create an MCP proxy via the sidecar HTTP API
Create a new tracewall run with a task contract
Create a run via the sidecar HTTP API
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Violates pattern:tool and pattern:response-shaper.
Parameter descriptions are minimal (20-50 chars vs 72-char baseline). No enums, ranges, or validation constraints. E.g., 'control_mode' accepts 'observe or enforce' but no enum declared; 'allowed_hosts' is an array with no item type or format specified.
Archive tools (sidecar.py, orchestration.py) are inferred from file paths, not visible in active MCP registration. Cannot verify actual tool definitions, schemas, or error handling.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Deny a pending approval via the sidecar HTTP API
Compute differences between two file snapshots
Evaluate an MCP request against policy
Generate a report for a tracewall run
Get JSON report for a run via the sidecar HTTP API
Get markdown report for a run via the sidecar HTTP API
Get run details via the sidecar HTTP API
Get pending approvals via the sidecar HTTP API
Record a shell command execution in a tracewall run
Record an event in a tracewall run
Create a snapshot of files in a directory
Stop a tracewall run and finalize recording
Stop a run via the sidecar HTTP API
Forward MCP requests through the sidecar proxy
Validate an MCP target URL for security constraints
Verify a tracewall run against its contract
Verify a run via the sidecar HTTP API
No error handling guidance. Tools lack recovery hints, retryability classification, or actionable error messages. E.g., 'verify_run' may fail but no guidance on why or what to do next.
Multiple tools combine concerns: create_run + record_event + stop_run form a workflow, not atomic operations. Violates pattern:tool (one thing per tool). No batch variants for agents calling in loops.