/sniper-compose -- Compose a Spawn Prompt from Persona Layers

You are executing the /sniper-compose command. Your job is to assemble a teammate spawn prompt by merging persona layers into the template. Follow every step below precisely.

The user's arguments are provided in: $ARGUMENTS

---

Step 0: Parse Arguments

Parse the following flags from $ARGUMENTS:

Flag	Required	Description	Example
--process	YES	Process persona layer name	architect
--technical	NO	Technical persona layer name (null if omitted)	backend
--cognitive	NO	Cognitive persona layer name (null if omitted)	security-first
--domain	NO	Domain context name from active domain pack	telephony
--name	YES	Display name for the teammate	"Backend Architect"
--ownership	NO	Ownership key from config.yaml	backend

If --process is missing, print an error:

ERROR: --process is required. This specifies the process persona layer.
Usage: /sniper-compose --process architect --technical backend --cognitive security-first --name "Backend Architect"
Available process layers: analyst, product-manager, architect, ux-designer, scrum-master, developer, qa-engineer

Then STOP.

If --name is missing, print an error:

ERROR: --name is required. This is the display name for the teammate.
Usage: /sniper-compose --process architect --technical backend --name "Backend Architect"

Then STOP.

If no arguments are provided at all, print a usage guide:

SNIPER Compose -- Assemble a spawn prompt from persona layers

Usage:
  /sniper-compose --process <layer> --name <name> [--technical <layer>] [--cognitive <layer>] [--domain <context>] [--ownership <key>]

Required:
  --process    Process persona layer (analyst, product-manager, architect, ux-designer, scrum-master, developer, qa-engineer)
  --name       Display name for the teammate (quoted if spaces)

Optional:
  --technical  Technical persona layer (backend, frontend, infrastructure, security, ai-ml, database, api-design)
  --cognitive  Cognitive persona layer (systems-thinker, security-first, performance-focused, user-empathetic, devils-advocate, mentor-explainer)
  --domain     Domain context from the active domain pack
  --ownership  Ownership key from config.yaml (backend, frontend, infrastructure, tests, docs)

Examples:
  /sniper-compose --process architect --technical backend --cognitive security-first --name "Backend Architect" --ownership backend
  /sniper-compose --process developer --technical frontend --cognitive user-empathetic --name "Frontend Dev" --ownership frontend
  /sniper-compose --process analyst --cognitive systems-thinker --name "Business Analyst"

Then STOP.

---

Step 1: Read the Template

Read the spawn prompt template from:

.sniper/spawn-prompts/_template.md

If it does not exist, print an error:

ERROR: Spawn prompt template not found at .sniper/spawn-prompts/_template.md
Run /sniper-init to set up the framework.

Then STOP.

Store the template content for merging in Step 4.

---

Step 2: Read Each Persona Layer

For each specified layer, read the corresponding file. Track which layers were found and which were not.

2a: Process Layer (required)

Read: .sniper/personas/process/{process_name}.md

If the file does not exist, check pack personas: .sniper/packs/*/personas/process/{process_name}.md

If found in a pack, use that file. If a pack persona has the same name as a framework persona, append the pack content after the framework content (pack extends framework).

If not found anywhere, print an error listing available process layers:

ERROR: Process persona '{process_name}' not found.
Available process layers:

Then list all .md files in .sniper/personas/process/ AND .sniper/packs/*/personas/process/ (without the extension). Then STOP.

2b: Technical Layer (optional)

If --technical was provided, read: .sniper/personas/technical/{technical_name}.md

If the file does not exist, print a warning:

WARNING: Technical persona '{technical_name}' not found. Skipping technical layer.
Available technical layers:

Then list all .md files in .sniper/personas/technical/. Set the technical layer content to:

No specific technical expertise assigned. Apply general engineering best practices.

2c: Cognitive Layer (optional)

If --cognitive was provided, read: .sniper/personas/cognitive/{cognitive_name}.md

If the file does not exist, print a warning and list available cognitive layers. Set the cognitive layer content to:

No specific cognitive style assigned. Apply balanced analytical thinking.

2d: Domain Layer (optional)

If --domain was provided:

1. Read .sniper/config.yaml to get the active domain_pack value
2. If domain_pack is null, print an error:

   ERROR: No domain pack is configured. Set domain_pack in .sniper/config.yaml or run /sniper-init.

   Set the domain layer content to No domain-specific context available.
3. If domain_pack is set, read: .sniper/domain-packs/{domain_pack}/context/{domain_name}.md
4. If the file does not exist, print a warning and list available contexts in that domain pack directory. Set the domain layer content to No domain-specific context available.

If --domain was NOT provided, set the domain layer content to:

No domain-specific context loaded. Refer to project documentation for domain knowledge.

---

Step 3: Read Ownership Rules

Read .sniper/config.yaml and extract the ownership section.

If --ownership was provided:

1. Look up the ownership key in the config (e.g., backend, frontend, infrastructure, tests, docs)
2. Extract the list of directory patterns for that key
3. Format them as a comma-separated string for the template

If --ownership was NOT provided:

1. Try to infer ownership from the --technical flag:
   • backend technical -> backend ownership
   • frontend technical -> frontend ownership
   • infrastructure technical -> infrastructure ownership
   • security technical -> backend ownership (security spans backend)
   • ai-ml technical -> backend ownership
   • database technical -> backend ownership
   • api-design technical -> backend ownership
2. If no technical layer was specified, set ownership to: "No specific ownership assigned -- coordinate with team lead"

Format the ownership value as a readable list, for example:

src/backend/, src/api/, src/services/, src/db/, src/workers/

---

Step 4: Merge Layers into Template

Take the template content from Step 1 and perform these replacements:

Placeholder	Replace With
{name}	The --name value
{process_layer}	Full content of the process persona file
{technical_layer}	Full content of the technical persona file (or default text)
{cognitive_layer}	Full content of the cognitive persona file (or default text)
{domain_layer}	Full content of the domain context file (or default text)
{memory_layer}	Memory layer content from Step 4b (or default text)
{ownership}	Formatted ownership directories from Step 3

The merged result is the complete spawn prompt.

---

Step 4b: Read Memory Layer

Read the project memory to inject relevant conventions, anti-patterns, and decisions into the spawn prompt.

4b-1: Check for Memory Files

Check if these files exist:

• .sniper/memory/conventions.yaml
• .sniper/memory/anti-patterns.yaml
• .sniper/memory/decisions.yaml

If none exist, set the memory layer content to:

No project memory available. This is either a new project or memory has not been initialized.

Skip to Step 5.

4b-2: Check for Workspace Memory

Read .sniper/config.yaml and check if workspace.enabled is true and workspace.workspace_path is set.

If workspace memory exists, also read:

• {workspace_path}/memory/conventions.yaml
• {workspace_path}/memory/anti-patterns.yaml
• {workspace_path}/memory/decisions.yaml

4b-3: Filter by Role

Determine the target agent's role from the --process flag value. Map process personas to agent roles:

• developer -> includes backend-engineer, frontend-engineer (based on --technical)
• architect -> includes architect
• analyst -> includes analyst
• qa-engineer -> includes qa-engineer
• retro-analyst -> includes retro-analyst

Filter all memory entries where applies_to includes the mapped role(s).

4b-4: Format Memory Layer

Format the filtered entries as a readable markdown section:

## Project Memory — Conventions You Must Follow

- **conv-001:** {rule}
  Example: {positive_example}
  NOT: {negative_example}

## Anti-Patterns — Do NOT Do These

- **ap-001 ({severity}):** {description}. Instead: {fix_pattern}

## Key Decisions

- **dec-001:** {title} — {decision}

If workspace conventions exist, label them with [WORKSPACE]:

- **conv-005 [WORKSPACE]:** {rule}

4b-5: Apply Token Budget

Read memory.token_budget from config (default: 2000).

If the formatted memory layer exceeds the token budget, truncate by priority:

1. High-severity anti-patterns (always include)
2. Conventions with enforcement: both or enforcement: review_gate
3. Active decisions
4. Medium-severity anti-patterns
5. Conventions with enforcement: spawn_prompt only
6. Low-severity anti-patterns

Add a note at the end if truncated:

Note: Memory layer truncated to fit token budget. Full memory available at .sniper/memory/

---

Step 5: Write the Composed Prompt

Generate a slug from the --name value:

• Lowercase
• Replace spaces with hyphens
• Remove special characters
• Example: "Backend Architect" -> backend-architect

Write the composed prompt to:

.sniper/spawn-prompts/{slug}.md

If a file already exists at that path, overwrite it silently (spawn prompts are regenerated as needed).

---

Step 6: Display Preview and Summary

Print a summary of what was composed:

============================================
  Spawn Prompt Composed
============================================

  Name:       {name}
  Saved to:   .sniper/spawn-prompts/{slug}.md

  Layers Used:
    Process:    {process_name} (.sniper/personas/process/{process_name}.md)
    Technical:  {technical_name or "none"}
    Cognitive:  {cognitive_name or "none"}
    Domain:     {domain_name or "none"}

  Ownership:   {ownership_dirs}

============================================

Then print a preview of the composed prompt. Show the FULL content of the generated file so the user can review it.

After the preview, print:

This spawn prompt is ready to use. It will be loaded when a teammate is created with this persona configuration.

---

IMPORTANT RULES

• Do NOT modify any persona layer source files -- they are read-only inputs.
• Do NOT modify the template file -- it is a read-only input.
• The composed output goes ONLY to .sniper/spawn-prompts/{slug}.md.
• If any required layer file is missing, STOP with a clear error. Do not generate placeholder content for required layers.
• For optional layers that are missing, use the default text specified above and continue.
• Always show the full preview so the user can verify before using the prompt.
• Preserve all markdown formatting from the persona layer files when merging.
• Do not add any extra content beyond what is in the template and layers.