Stack-Aware Intelligence System

Auto-detects project tech stack and loads domain-relevant context for reviews, audits, planning, and work.

Architecture (4 Layers)

Layer 0: Context Router — computeContextManifest() decides WHAT to load
Layer 1: Detection Engine — detectStack() scans manifests for evidence
Layer 2: Knowledge Skills — 20 reference docs (languages, frameworks, databases, libraries, patterns)
Layer 3: Enforcement Agents — 13 specialist reviewers with unique finding prefixes

How It Works

1. Detection: detectStack() scans manifest files (pyproject.toml, package.json, Cargo.toml, composer.json) to identify languages, frameworks, databases, and libraries with a confidence score (0.0-1.0).

1. Routing: computeContextManifest() classifies changed files into domains (backend, frontend, database, testing, infra, docs), then selects which reference docs and agents to load based on the detected stack and task type.

1. Loading: Selected reference docs are passed as Read() directives to workers and reviewers. Only domain-relevant docs are loaded (e.g., a frontend-only PR won't load Python/SQLAlchemy skills).

1. Enforcement: Specialist agents (python-reviewer, fastapi-reviewer, design-system-compliance-reviewer, etc.) are conditionally summoned when their stack is detected AND domain matches, producing findings with unique prefixes (PY-, FAPI-, etc.).

Key Functions

detectStack(repoRoot)

See detection.md for the full algorithm.

Input: Repository root path Output: { primary_language, languages[], frameworks[], databases[], libraries[], tooling[], confidence, evidence_files[] }

computeContextManifest(task_type, file_scope, detected_stack, task_description)

See context-router.md for the full algorithm.

Input: Workflow type (review/audit/plan/work/forge), changed files, detected stack, task description Output: { domains, skills_to_load[], skills_excluded{}, agents_selected[], loading_strategy }

Supported Stacks

See stack-registry.md for all supported languages, frameworks, databases, and libraries.

Configuration

# talisman.yml
stack_awareness:
  enabled: true                  # Master toggle (default: true)
  confidence_threshold: 0.5      # Min confidence to activate specialists
  max_stack_ashes: 3             # Max specialist Ashes per review

  # Override detected stack (monorepos, detection failures)
  # override:
  #   primary_language: python
  #   frameworks: [fastapi, sqlalchemy]

  # Custom project-specific rules
  # custom_rules:
  #   - path: ".claude/skills/my-patterns/SKILL.md"
  #     domains: [backend]
  #     workflows: [review, work]
  #     stacks: [python]

References

• detection.md — Stack detection algorithm
• stack-registry.md — Supported stacks registry
• context-router.md — Smart context loading algorithm
• languages/ — Language profiles (Python, TypeScript, Rust, PHP)
• frameworks/ — Framework checklists (FastAPI, Django, Laravel, SQLAlchemy, React, Vue.js, Next.js, Vite)
• databases/ — Database patterns (PostgreSQL, MySQL)
• libraries/ — Library patterns (Pydantic, Returns, Dishka)
• patterns/ — Cross-cutting patterns (TDD, DDD, DI)