MCP Prompts

SYNAPSEED provides 5 prompt templates that guide the LLM through complex multi-step workflows using the available tools and resources.

describe_architecture

Analyze and describe the project architecture using SYNAPSEED's semantic understanding.

Arguments:

Name	Required	Description
path	No	Directory to analyze (default: project root)
depth	No	overview, detailed (default), or deep

Workflow:

1. hoist to index the project
2. Read synapseed://status for project state
3. Read synapseed://dna for configuration
4. lookup for key types (detailed/deep modes)
5. blame on key files (deep mode only)

---

fix_build_errors

Diagnose and fix current build errors using the shadow compiler.

Arguments:

Name	Required	Description
auto_fix	No	true to auto-apply, false (default) to ask first

Workflow:

1. Read synapseed://diagnostics/active
2. diagnostics for full error list
3. Group by file and severity
4. quickfix for MachineApplicable suggestions
5. Re-check diagnostics to verify

---

explain_evolution

Analyze why code looks the way it does by tracing its git evolution.

Arguments:

Name	Required	Description
file	Yes	File path to analyze
start_line	No	Start line of region
end_line	No	End line of region

Workflow:

1. hoist for project structure
2. lookup for symbols in the file
3. analyze for churn and patterns
4. blame for detailed blame

Output: Timeline, complexity narrative, risk assessment, recommendations.

---

security_audit

Perform a security audit of the project.

Arguments:

Name	Required	Description
scope	No	quick, standard (default), or full

Workflow (standard):

1. Read synapseed://security/policy
2. Read synapseed://status
3. hoist to find all files
4. scan on each config file
5. check on common project commands
6. diagnose for overall health

Output: DLP findings, command policy evaluation, risk assessment.

---

optimize_hotspots

Analyze runtime performance hotspots from OTLP telemetry data.

Arguments:

Name	Required	Description
threshold	No	Minimum avg_duration_ms to flag (default: 100)

Workflow:

1. Read synapseed://telemetry/hotspots
2. Filter by threshold
3. lookup for each hot function
4. analyze for churn context
5. diagnostics for compiler warnings

Output: Hotspot report, optimization suggestions, priority-ordered action plan.