icdev 1.0.0__py3-none-any.whl

icdev/data/args/llm_config.yaml

@@ -0,0 +1,495 @@
+# [TEMPLATE: CUI // SP-CTI]
+# ICDEV LLM Provider Configuration
+# Controls which models are used for each function across the platform.
+# Single source of truth — replaces scattered hardcoded model IDs.
+
+# -----------------------------------------------------------------------
+# Provider Definitions
+# -----------------------------------------------------------------------
+providers:
+  bedrock:
+    type: bedrock
+    region: ${AWS_DEFAULT_REGION:-us-gov-west-1}
+    # Uses boto3 credentials (IAM role, env vars, or ~/.aws/credentials)
+
+  anthropic:
+    type: anthropic
+    api_key_env: ANTHROPIC_API_KEY
+    base_url: https://api.anthropic.com
+
+  openai:
+    type: openai_compatible
+    api_key_env: OPENAI_API_KEY
+    base_url: https://api.openai.com/v1
+
+  ollama:
+    type: ollama
+    base_url: ${OLLAMA_BASE_URL:-http://localhost:11434}
+
+  gemini:
+    type: gemini
+    api_key_env: GOOGLE_API_KEY
+
+  vllm:
+    type: openai_compatible
+    base_url: ${VLLM_BASE_URL:-http://localhost:8000/v1}
+    api_key_env: VLLM_API_KEY
+
+  ibm_watsonx:
+    type: ibm_watsonx
+    api_key_env: IBM_CLOUD_API_KEY
+    project_id: ${IBM_WATSONX_PROJECT_ID:-}
+    url: ${IBM_WATSONX_URL:-https://us-south.ml.cloud.ibm.com}
+
+# -----------------------------------------------------------------------
+# Model Registry
+# -----------------------------------------------------------------------
+models:
+  # -- Anthropic via Bedrock --
+  claude-opus:
+    provider: bedrock
+    model_id: anthropic.claude-opus-4-6-20260215-v1:0
+    max_output_tokens: 128000
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.005, output_per_1k: 0.025 }
+
+  claude-sonnet:
+    provider: bedrock
+    model_id: anthropic.claude-sonnet-4-5-20250929-v1:0
+    max_output_tokens: 64000
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.003, output_per_1k: 0.015 }
+
+  claude-haiku:
+    provider: bedrock
+    model_id: anthropic.claude-3-5-sonnet-20241022-v2:0
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: false
+    pricing: { input_per_1k: 0.003, output_per_1k: 0.015 }
+
+  # -- Anthropic Direct API --
+  claude-opus-api:
+    provider: anthropic
+    model_id: claude-opus-4-6-20260205
+    max_output_tokens: 128000
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    pricing: { input_per_1k: 0.005, output_per_1k: 0.025 }
+
+  claude-sonnet-api:
+    provider: anthropic
+    model_id: claude-sonnet-4-5-20250929
+    max_output_tokens: 64000
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    pricing: { input_per_1k: 0.003, output_per_1k: 0.015 }
+
+  # -- OpenAI --
+  gpt-4o:
+    provider: openai
+    model_id: gpt-4o
+    max_output_tokens: 16384
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.0025, output_per_1k: 0.010 }
+
+  gpt-4o-mini:
+    provider: openai
+    model_id: gpt-4o-mini
+    max_output_tokens: 16384
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: true
+    pricing: { input_per_1k: 0.00015, output_per_1k: 0.0006 }
+
+  o3:
+    provider: openai
+    model_id: o3
+    max_output_tokens: 100000
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    pricing: { input_per_1k: 0.010, output_per_1k: 0.040 }
+
+  # -- Google Gemini --
+  gemini-2.5-pro:
+    provider: gemini
+    model_id: gemini-2.5-pro-preview-05-06
+    max_output_tokens: 65536
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.00125, output_per_1k: 0.010 }
+
+  gemini-2.5-flash:
+    provider: gemini
+    model_id: gemini-2.5-flash-preview-04-17
+    max_output_tokens: 65536
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.00015, output_per_1k: 0.0006 }
+
+  gemini-2.0-flash:
+    provider: gemini
+    model_id: gemini-2.0-flash
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: true
+    supports_vision: true
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  # -- Local Models (Ollama) --
+  llama-local:
+    provider: ollama
+    model_id: llama3.3:70b
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: false
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  codestral-local:
+    provider: ollama
+    model_id: codestral:22b
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: false
+    supports_structured_output: false
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  deepseek-local:
+    provider: ollama
+    model_id: deepseek-r1:70b
+    max_output_tokens: 8192
+    supports_thinking: true
+    supports_tools: false
+    supports_structured_output: false
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  qwen3-local:
+    provider: ollama
+    model_id: qwen3:latest
+    max_output_tokens: 32768
+    supports_thinking: true
+    supports_tools: true
+    supports_structured_output: true
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  # -- Vision Models (Local) --
+  gemma3-local:
+    provider: ollama
+    model_id: gemma3:latest
+    max_output_tokens: 4096
+    supports_thinking: false
+    supports_tools: false
+    supports_structured_output: false
+    supports_vision: true
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  llava-local:
+    provider: ollama
+    model_id: llava:13b
+    max_output_tokens: 4096
+    supports_thinking: false
+    supports_tools: false
+    supports_structured_output: false
+    supports_vision: true
+    pricing: { input_per_1k: 0.0, output_per_1k: 0.0 }
+
+  # IBM watsonx.ai models (D238)
+  granite-code:
+    provider: ibm_watsonx
+    model_id: ibm/granite-34b-code-instruct
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: false
+    supports_structured_output: false
+    pricing:
+      input_per_1k: 0.0
+      output_per_1k: 0.0
+  granite-chat:
+    provider: ibm_watsonx
+    model_id: ibm/granite-3.1-8b-instruct
+    max_output_tokens: 8192
+    supports_thinking: false
+    supports_tools: true
+    supports_structured_output: false
+    pricing:
+      input_per_1k: 0.0
+      output_per_1k: 0.0
+
+# -----------------------------------------------------------------------
+# Function -> Model Routing
+# -----------------------------------------------------------------------
+# Maps each ICDEV function to a model with fallback chain.
+# First available model in the chain is used.
+routing:
+  agent_architect:
+    chain: [claude-opus, o3, gemini-2.5-pro, gpt-4o, qwen3-local, llama-local]
+    effort: max
+
+  agent_orchestrator:
+    chain: [claude-opus, gemini-2.5-pro, gpt-4o, qwen3-local, llama-local]
+    effort: high
+
+  agent_builder:
+    chain: [claude-opus, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: max
+
+  agent_compliance:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: high
+
+  agent_security:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: high
+
+  agent_infra:
+    chain: [claude-sonnet, gpt-4o-mini, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: medium
+
+  agent_knowledge:
+    chain: [claude-sonnet, gpt-4o-mini, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: medium
+
+  agent_monitor:
+    chain: [claude-haiku, gpt-4o-mini, gemini-2.0-flash, qwen3-local, llama-local]
+    effort: low
+
+  task_decomposition:
+    chain: [claude-opus, o3, gemini-2.5-pro, gpt-4o, qwen3-local]
+    effort: high
+
+  collaboration:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: medium
+
+  nlq_sql:
+    chain: [gpt-4o-mini, gemini-2.0-flash, claude-haiku, qwen3-local, llama-local]
+    effort: low
+
+  code_generation:
+    chain: [claude-sonnet, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: high
+
+  saas_proxy:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local]
+    effort: medium
+
+  child_app:
+    chain: [claude-sonnet, gemini-2.5-pro, gpt-4o, qwen3-local]
+    effort: medium
+
+  screenshot_validation:
+    chain: [claude-opus, claude-sonnet, gpt-4o, gemini-2.5-flash, gemma3-local, llava-local]
+    effort: medium
+
+  document_vision:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, gemma3-local, llava-local]
+    effort: medium
+
+  ui_analysis:
+    chain: [claude-opus, claude-sonnet, gpt-4o, gemini-2.5-pro, gemma3-local, llava-local]
+    effort: high
+
+  diagram_extraction:
+    chain: [claude-opus, claude-sonnet, gpt-4o, gemini-2.5-pro, gemma3-local, llava-local]
+    effort: high
+
+  compliance_diagram:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, gemma3-local, llava-local]
+    effort: medium
+
+  attachment_analysis:
+    chain: [claude-sonnet, gpt-4o-mini, gemini-2.0-flash, gemma3-local, llava-local]
+    effort: low
+
+  intake_persona_response:
+    chain: [claude-sonnet, claude-opus, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: medium
+
+  bdd_preview:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, codestral-local]
+    effort: low
+
+  narrative_generation:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: high
+
+  compliance_export:
+    chain: [claude-sonnet, gpt-4o-mini, gemini-2.0-flash, qwen3-local, llama-local]
+    effort: medium
+
+  # Cross-Language Translation (Phase 43)
+  code_translation:
+    chain: [claude-opus, claude-sonnet, o3, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: max
+
+  code_translation_repair:
+    chain: [claude-sonnet, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: high
+
+  code_translation_review:
+    chain: [claude-sonnet, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: high
+
+  test_translation:
+    chain: [claude-sonnet, gemini-2.5-pro, gpt-4o, qwen3-local, codestral-local]
+    effort: high
+
+  dependency_suggestion:
+    chain: [claude-haiku, gpt-4o-mini, gemini-2.0-flash, qwen3-local, codestral-local]
+    effort: low
+
+  default:
+    chain: [claude-sonnet, gpt-4o, gemini-2.5-flash, qwen3-local, llama-local]
+    effort: medium
+
+# -----------------------------------------------------------------------
+# Per-Agent Effort Defaults (Enhancement #4 — used by agent_executor)
+# -----------------------------------------------------------------------
+agent_effort_defaults:
+  orchestrator-agent: high
+  architect-agent: max
+  builder-agent: max
+  compliance-agent: high
+  security-agent: high
+  infra-agent: medium
+  knowledge-agent: medium
+  monitor-agent: low
+  mbse-agent: high
+  modernization-agent: medium
+  requirements-agent: medium
+  supply-chain-agent: medium
+  simulation-agent: high
+  devsecops-agent: high
+
+# -----------------------------------------------------------------------
+# Embedding Configuration
+# -----------------------------------------------------------------------
+embeddings:
+  default_chain: [openai-embed, gemini-embed, nomic-embed-local]
+
+  models:
+    openai-embed:
+      provider: openai
+      model_id: text-embedding-3-small
+      dimensions: 1536
+      pricing: { input_per_1k: 0.00002 }
+
+    gemini-embed:
+      provider: gemini
+      model_id: text-embedding-004
+      dimensions: 768
+      pricing: { input_per_1k: 0.0 }
+
+    titan-embed:
+      provider: bedrock
+      model_id: amazon.titan-embed-text-v2:0
+      dimensions: 1024
+      pricing: { input_per_1k: 0.00002 }
+
+    nomic-embed-local:
+      provider: ollama
+      model_id: nomic-embed-text
+      dimensions: 768
+      pricing: { input_per_1k: 0.0 }
+
+    # Azure OpenAI embedding (D238)
+    azure-embed:
+      provider: azure_openai
+      model_id: text-embedding-ada-002
+      dimensions: 1536
+      pricing:
+        input_per_1k: 0.0001
+    # OCI GenAI embedding (D238)
+    oci-embed:
+      provider: oci_genai
+      model_id: cohere.embed-english-v3.0
+      dimensions: 1024
+      pricing:
+        input_per_1k: 0.0001
+    # IBM watsonx.ai embedding (D238)
+    ibm-slate-embed:
+      provider: ibm_watsonx
+      model_id: ibm/slate-125m-english-rtrvr-v2
+      dimensions: 768
+      pricing:
+        input_per_1k: 0.0
+
+# -----------------------------------------------------------------------
+# Global Settings
+# -----------------------------------------------------------------------
+settings:
+  max_retries: 5
+  base_retry_delay_seconds: 1.0
+  max_retry_delay_seconds: 30.0
+  availability_cache_ttl_seconds: 1800
+  prefer_local: false    # Two-tier handles local/cloud split — do not set true (blocks Claude tier2)
+
+# -----------------------------------------------------------------------
+# Two-Tier Routing (qwen3 worker → Claude planner/reviewer)
+# -----------------------------------------------------------------------
+# Claude is the brain. qwen3 handles drafting, research, and scanning.
+#
+# planner_functions : Claude directly — no qwen3 pre-step (architecture,
+#                     orchestration, high-stakes planning)
+# worker_functions  : qwen3 drafts a COMPACT structured response, then
+#                     Claude verifies/corrects and returns the final answer.
+#                     qwen3 output is intentionally short (reduces Claude's
+#                     input tokens vs. Claude doing the full task alone).
+# scanner_functions : qwen3 only — log analysis, health checks, simple
+#                     pattern matching where review adds no value.
+#
+# Set enabled: false to fall back to normal chain-based routing.
+two_tier:
+  enabled: true
+  tier1_model: qwen3-local    # Worker — research, draft, scan (free/local)
+  tier2_model: claude-sonnet-api  # Planner/Reviewer — Anthropic API (not Bedrock)
+
+  planner_functions:
+    - agent_architect
+    - agent_orchestrator
+    - task_decomposition
+    - child_app
+    - saas_proxy
+
+  worker_functions:
+    - agent_builder
+    - agent_compliance
+    - agent_security
+    - agent_infra
+    - agent_knowledge
+    - code_generation
+    - code_translation
+    - code_translation_repair
+    - code_translation_review
+    - test_translation
+    - bdd_preview
+    - nlq_sql
+    - narrative_generation
+    - compliance_export
+    - intake_persona_response
+    - collaboration
+    - dependency_suggestion
+    - default
+
+  scanner_functions:
+    - agent_monitor
+    - attachment_analysis

icdev/data/args/maintenance_config.yaml

@@ -0,0 +1,55 @@
+# Maintenance Audit Configuration
+# Controls dependency scanning, vulnerability checking, SLA enforcement, and remediation
+
+# SLA (Service Level Agreement) thresholds for vulnerability remediation
+sla:
+  critical_hours: 48       # 2 days — critical vulnerabilities
+  high_hours: 168          # 7 days — high vulnerabilities
+  medium_hours: 720        # 30 days — medium vulnerabilities
+  low_hours: 2160          # 90 days — low vulnerabilities
+
+# Dependency staleness thresholds
+staleness:
+  warning_days: 90         # Flag as warning if dependency is >90 days behind latest
+  critical_days: 180       # Flag as critical if dependency is >180 days behind latest
+  max_acceptable_days: 365 # Hard limit — anything older is a finding
+
+# Maintenance score calculation
+scoring:
+  overdue_critical_penalty: 15   # Points deducted per overdue critical SLA
+  overdue_high_penalty: 10       # Points deducted per overdue high SLA
+  overdue_medium_penalty: 5      # Points deducted per overdue medium SLA
+  overdue_low_penalty: 2         # Points deducted per overdue low SLA
+  staleness_penalty_per_day: 0.1 # Points deducted per day of avg staleness above baseline
+  staleness_baseline_days: 90    # No staleness penalty below this threshold
+
+# Remediation engine settings
+remediation:
+  auto_remediate: true                  # Enable automatic remediation
+  auto_remediate_max_severity: "medium" # Auto-fix medium and below without confirmation
+  require_tests_pass: true              # Require tests to pass before marking as applied
+  require_lint_pass: true               # Require linting to pass before marking as applied
+  branch_prefix: "remediation/"         # Git branch prefix for remediation branches
+  commit_prefix: "[MAINT]"             # Commit message prefix for remediation commits
+  max_concurrent_remediations: 5        # Max simultaneous remediation branches
+
+# Advisory/vulnerability data sources
+advisory_sources:
+  - "tool-native"    # Use language-native tools (pip-audit, npm audit, cargo-audit, etc.)
+  - "osv"            # OSV.dev API (unified, open-source vulnerability database)
+
+# Reporting settings
+reporting:
+  trend_lookback_days: 90    # How far back to look for trend analysis
+  include_resolved: false    # Include resolved vulnerabilities in reports
+  report_format: "markdown"  # Report format (markdown)
+  cui_markings: true         # Apply CUI markings to reports
+
+# Supported package registries for version lookups (used by dependency_scanner)
+registries:
+  pypi: "https://pypi.org/pypi/{package}/json"
+  npm: "https://registry.npmjs.org/{package}"
+  maven: "https://search.maven.org/solrsearch/select?q=g:{group}+AND+a:{artifact}&rows=1&wt=json"
+  crates: "https://crates.io/api/v1/crates/{crate}"
+  go: "https://proxy.golang.org/{module}/@latest"
+  nuget: "https://api.nuget.org/v3-flatcontainer/{package}/index.json"

icdev/data/args/memory_config.yaml

@@ -0,0 +1,83 @@
+# [TEMPLATE: CUI // SP-CTI]
+# Memory System Configuration — time-decay, dedup, auto-capture, maintenance
+# (D147, D179-D182)
+---
+
+time_decay:
+  # Exponential decay formula: decay_factor = 2^(-(age_days / half_life_days))
+  # Per-type half-lives (in days)
+  half_lives:
+    fact: 90          # Long-lived — organizational knowledge
+    preference: 180   # Very long-lived — user/system preferences rarely change
+    event: 7          # Short-lived — session events lose relevance quickly
+    insight: 30       # Medium — insights remain useful for ~1 month
+    task: 14          # Short-medium — task context relevant during sprint
+    relationship: 120 # Long-lived — people/system relationships persist
+    thinking: 3       # Very short — reasoning traces decay rapidly (D182)
+
+  # Default half-life for unknown memory types
+  default_half_life: 30
+
+  # Minimum decay factor (prevents ancient entries from being completely zeroed)
+  min_decay_factor: 0.01
+
+  # Scoring weights when time-decay is enabled (must sum to 1.0)
+  weights:
+    relevance: 0.60    # BM25 + semantic hybrid score
+    recency: 0.25      # Time decay factor
+    importance: 0.15   # Normalized importance (importance / 10.0)
+
+  # Override: if importance >= this threshold, halve the decay effect
+  # (very important memories resist decay)
+  importance_decay_resistance_threshold: 8
+
+# Content deduplication (D179)
+deduplication:
+  enabled: true
+  hash_algorithm: "sha256"    # SHA-256 content hash before INSERT
+
+# Auto-capture settings (D181)
+auto_capture:
+  enabled: true
+  buffer_flush_threshold: 100   # Flush when buffer exceeds this count
+  default_importance: 3         # Importance for auto-captured entries
+  max_content_length: 2000      # Truncate captured content to this length
+
+# Maintenance pipeline settings (D179-D182)
+maintenance:
+  enabled: true
+  prune_stale_days: 180         # Prune low-importance entries older than this
+  prune_min_importance: 3       # Only prune entries with importance <= this
+  prune_types:                  # Only prune these memory types
+    - event
+    - thinking
+  embed_batch_size: 20          # Batch size for embedding generation
+  backup_before_prune: true     # Backup memory.db before pruning
+
+# History compression (Phase 44 — D271-D274)
+history_compression:
+  enabled: true                   # Opt-in per context
+  budget_allocation:
+    current_topic: 0.50           # 50% of budget for current topic
+    historical_topics: 0.30       # 30% for historical summaries
+    bulk: 0.20                    # 20% for bulk merged content
+  topic_detection:
+    time_gap_minutes: 30          # Gap > 30min = new topic boundary
+    keyword_shift_threshold: 0.60 # Keyword overlap < 40% = new topic
+  summarization:
+    llm_function: "history_summarize"
+    fallback: "truncation"        # Use truncation when LLM unavailable
+    max_summary_tokens: 200       # Max tokens per topic summary
+
+# Memory consolidation (Phase 44 — D276)
+consolidation:
+  enabled: true
+  similarity_threshold: 0.75     # Minimum similarity to trigger consolidation
+  use_llm: true                  # Use LLM for merge/replace decisions
+  llm_function: "memory_consolidation"
+  max_candidates: 5              # Max similar entries to consider
+  batch_size: 50                 # Batch size for consolidate_all()
+  actions:
+    jaccard_skip: 0.90            # Jaccard >= 0.90 → SKIP (near-duplicate)
+    jaccard_replace: 0.80         # Jaccard >= 0.80 → REPLACE
+    jaccard_keep: 0.75            # Jaccard >= 0.75 → KEEP_SEPARATE