skill-injection-protocol

name	skill-injection-protocol
description	Use when resolving language-specific skills for a workflow. Triggers on multi-language projects or when language-specific conventions are needed before implementation.

Skill Injection Protocol (Orchestrator)

Resolve language-specific skills based on project tech stack and profile configuration. The orchestrator executes this protocol ONCE per workflow and caches the result as language_skills_prompt in workflow-state.yaml.

Output

A language_skills_prompt string value:

• Skills found: formatted prompt fragment (see Output Format)
• No skills: empty string ""

Steps

1. Read Profile Configuration

Read .agentic.settings.json from the IDE directory. Extract the profiles array.

{
  "profiles": [
    { "name": "typescript", "detect": ["typescript", "ts", "node", "bun", "deno"], "skills": ["typescript-engineer"] },
    { "name": "python", "detect": ["python", "py", "pip", "uv", "poetry", "conda"], "skills": ["python-engineer"] }
  ]
}

If profiles is missing or empty → language_skills_prompt: "", stop.

2. Read Tech Stack

Read technical-context.md from the workflow output path. Extract tech_stack from YAML frontmatter.

---
tech_stack: [typescript, react, node]
---

If technical-context.md doesn't exist or tech_stack is missing → language_skills_prompt: "", stop.

3. Match & Deduplicate

For each tech_stack entry, search all profiles' detect arrays (case-insensitive). Collect matched profiles' skills. Deduplicate.

Example: tech_stack [typescript, react, node] → all match typescript profile → skills: ["typescript-engineer"]

4. Apply Overrides

Check skillOverrides in settings:

Override	Example	Effect
Replace	{"typescript-engineer": "custom-ts"}	Load custom-ts instead
Remove	{"typescript-engineer": "_remove_"}	Don't load that skill

5. Handle Unmatched

If tech_stack entries have no matching profile:

• Auto mode: Log in decision-log.md as LOW_CONFIDENCE
• Interactive mode: Ask user

If no skills matched → language_skills_prompt: "", stop.

6. Build Output

Set language_skills_prompt to:

---

# Language Skills

Load these language-specific skills before proceeding:
- Skill(skill="{skill-1}")
- Skill(skill="{skill-2}")

Confirm: "Language skills loaded: {skill-1}, {skill-2}"

---

Cache in workflow-state.yaml:

language_skills_prompt: |
  ---

  # Language Skills

  Load these language-specific skills before proceeding:
  - Skill(skill="{skill-1}")
  - Skill(skill="{skill-2}")

  Confirm: "Language skills loaded: {skill-1}, {skill-2}"

  ---

Common Mistakes

Mistake	Fix
Agent tries to execute this protocol	Only orchestrator executes this. Agents receive {language_skills_prompt} in their Task() prompt.
Resolving per-step instead of once	Resolve once at the workflow resolution point, cache in workflow-state.yaml.
Wrong skill name prefix	Use short names: typescript-engineer, not agentic-skill-typescript-engineer