AgentLeak
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| privacy_preflightA | Check this project for privacy problems before shipping, and report which findings are NEW since the last check. Call this after writing or changing code that touches user data, secrets, logging, or outbound calls. Optionally pass a trace of the run you just made to score it across all eight channels as well. |
| privacy_scan_codeB | Scan source code for hardcoded secrets, PII written to logs, and sensitive values sent to third parties. Use when you only need the current state and not the change since last time. |
| privacy_check_traceA | Score one agent run for privacy leakage across tool calls, tool responses, shared memory, inter-agent messages, logs, generated files and the final answer. Use when you have captured a run and want to know whether data escaped on a channel nobody reads. |
| privacy_redactA | Remove sensitive values from text. Use before writing something to a log, an issue, a commit message, or any other place a secret should not end up. Styles: placeholder, masked, asterisk, category, hash, remove. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 4 tools
The four tools target distinct concerns: delta-aware project check, static code scan, single-run trace scoring, and text redaction. However, privacy_preflight and privacy_scan_code overlap substantially (preflight is essentially scan plus a delta and optional trace scoring), so an agent must rely on the descriptions to choose correctly rather than on clearly separate purposes.
All names share the privacy_ prefix and are snake_case, giving a strong, predictable family. Minor deviation: scan_code and check_trace follow a verb_noun shape while preflight and redact are verb/noun-only, so the action pattern isn't perfectly uniform.
Four tools is a tight, well-scoped surface for a privacy-auditing server, and each one covers a distinct mode of use (pre-ship delta check, static scan, trace scoring, redaction). No tool looks redundant or bolted on.
The lifecycle is largely covered: detect (scan), detect-and-diff (preflight), deep channel analysis (trace), and remediate (redact). Gaps exist around retrieving or listing prior findings, configuring policy/severity thresholds, and verifiable proof of remediation, but these are workaroundable.