terraphim_orchestrator 1.16.34

# orchestrator.example.toml -- AI Dark Factory Agent Fleet
#
# Copy to orchestrator.toml and adjust paths for your environment.
# Full 13-agent fleet with persona-specific knowledge graphs,
# terraphim-agent pre-checks, and disciplined engineering skill chains.

working_dir = "/opt/ai-dark-factory/workspace"
persona_data_dir = "/opt/ai-dark-factory/personas"
skill_data_dir = "/opt/ai-dark-factory/skills"
role_config_path = "/opt/ai-dark-factory/persona_roles_config.json"

[nightwatch]
eval_interval_secs = 300
minor_threshold = 0.10
moderate_threshold = 0.20
severe_threshold = 0.40
critical_threshold = 0.70

[compound_review]
schedule = "0 2 * * *"
max_duration_secs = 1800
repo_path = "/opt/ai-dark-factory/workspace"
create_prs = false

# =============================================================================
# ROUTING (telemetry-aware model selection)
# =============================================================================
# Uncomment to enable KG-driven and telemetry-aware model routing.
# Requires taxonomy markdown files in the taxonomy_path directory.
# When use_routing_engine is true, the control-plane routing engine
# combines KG routing, keyword routing, provider health, budget pressure,
# and live telemetry signals for model selection.
#
# [routing]
# taxonomy_path = "/opt/ai-dark-factory/kg/routing"
# probe_ttl_secs = 300
# probe_on_startup = true
# use_routing_engine = false

# =============================================================================
# SAFETY LAYER (always running)
# =============================================================================

[[agents]]
name = "security-sentinel"
layer = "Safety"
cli_tool = "claude"
persona = "vigil"
terraphim_role = "Vigil Security"
skill_chain = ["disciplined-verification", "disciplined-validation"]
task = """Continuously scan for CVEs and security vulnerabilities in dependencies.
Run cargo audit, check for OWASP Top 10 issues, and report findings.

## Domain Knowledge Search
When you need security context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Vigil Security"

## MANDATORY: Post verdict to Gitea
Post your security verdict to the relevant Gitea issue.
- PASS if no critical security issues found
- FAIL if critical vulnerabilities or issues found

Post verdict with @adf:merge-coordinator trigger.
Replace ISSUE_NUMBER with the issue from mention context, or the default tracking issue if no mention context.
"""
capabilities = ["security", "vulnerability-scanning", "supply-chain"]
max_memory_bytes = 2147483648

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "CVE vulnerability dependency audit supply chain" --role "Vigil Security" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "drift-detector"
layer = "Safety"
cli_tool = "claude"
persona = "echo"
terraphim_role = "Echo QA"
skill_chain = ["disciplined-verification", "disciplined-validation"]
task = """Monitor for configuration drift, test failures, and quality regressions.
Compare current state against known-good baselines and report deviations.

## Domain Knowledge Search
When you need QA context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Echo QA"
"""
capabilities = ["drift-detection", "quality-monitoring", "regression-detection"]
max_memory_bytes = 2147483648

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "regression drift quality baseline test failure" --role "Echo QA" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

# =============================================================================
# CORE LAYER (scheduled)
# =============================================================================

[[agents]]
name = "test-guardian"
layer = "Core"
cli_tool = "claude"
persona = "echo"
terraphim_role = "Echo QA"
skill_chain = ["disciplined-verification", "disciplined-validation", "testing", "acceptance-testing"]
schedule = "15 */8 * * *"
capabilities = ["testing", "coverage", "quality"]
max_cpu_seconds = 3600
grace_period_secs = 30
task = """Run comprehensive test suite and quality analysis:
1. cd to the workspace
2. cargo test --workspace 2>&1 | tee /tmp/test-guardian-results.txt
3. Identify failing or flaky tests
4. Check for untested critical code paths
5. Run: cargo clippy -- -D warnings 2>&1
6. Summarise findings as a test report

## MANDATORY: Post verdict to Gitea
Post your verdict to the relevant Gitea issue. Use:
- PASS if all tests pass and no critical quality issues
- FAIL if tests fail or critical quality issues found

Post verdict with @adf:merge-coordinator trigger.
Replace ISSUE_NUMBER with the issue from mention context, or the default tracking issue if no mention context.

## Domain Knowledge Search
When you need QA context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Echo QA"
"""

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "test quality regression flaky coverage" --role "Echo QA" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "implementation-swarm"
layer = "Core"
cli_tool = "claude"
persona = "ferrox"
terraphim_role = "Ferrox Rust"
skill_chain = ["disciplined-research", "disciplined-design", "disciplined-implementation", "disciplined-verification", "disciplined-validation"]
task = """Implement the highest-priority unblocked Gitea issue:
1. Run: gitea-robot ready --owner terraphim --repo terraphim-ai
2. Pick the highest PageRank unblocked issue (is_blocked=false)
3. Read the full issue details
4. Create branch: git checkout -b task/<IDX>-<short-title>
5. Implement following the issue spec
6. Run: cargo check && cargo clippy -- -D warnings && cargo fmt --all -- --check
7. Run: cargo test (for the affected package)
8. Quality check the changes for correctness, security, and idiomatic Rust
9. If quality check passes, commit with Refs #<IDX> and push branch:
   git add -A && git commit -m "feat: implement #<IDX> - <short-title> Refs #<IDX>" && git push -u origin task/<IDX>-<short-title>
10. Create PR:
   gitea-robot create-pull --owner terraphim --repo terraphim-ai --title "Fix #<IDX>: <short-title>" --base main --head task/<IDX>-<short-title>
11. Post comment to trigger review chain:
   gitea-robot comment --owner terraphim --repo terraphim-ai --index <IDX> --body "Implementation complete. Quality check passed. PR created from branch task/<IDX>-<short-title>.
   @adf:quality-coordinator please review this implementation for issue #<IDX>"
12. DO NOT close the issue. The merge-coordinator will close it after all reviews pass.

If quality check fails, fix the issues and re-run steps 6-8.

## Domain Knowledge Search
When you need Rust context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Ferrox Rust"
"""
schedule = "0 1 * * *"
capabilities = ["implementation", "rust-development", "tdd"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "implementation async ownership error handling" --role "Ferrox Rust" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "product-development"
layer = "Core"
cli_tool = "claude"
persona = "ferrox"
terraphim_role = "Ferrox Rust"
skill_chain = ["disciplined-research", "disciplined-design", "disciplined-implementation", "disciplined-verification", "disciplined-validation"]
task = """Develop new product features based on roadmap priorities.
Focus on user-facing functionality, integration points, and API design.

## Domain Knowledge Search
When you need Rust context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Ferrox Rust"
"""
schedule = "0 2 * * *"
capabilities = ["product-development", "feature-implementation", "api-design"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "feature api service integration" --role "Ferrox Rust" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "meta-coordinator"
layer = "Core"
cli_tool = "claude"
persona = "carthos"
terraphim_role = "Carthos Architecture"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Coordinate agent activities, review architectural decisions, ensure
bounded contexts are respected, and maintain system coherence.

## Domain Knowledge Search
When you need architecture context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Carthos Architecture"
"""
schedule = "0 0 * * *"
capabilities = ["coordination", "architecture-review", "system-design"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "bounded context architecture decision API design" --role "Carthos Architecture" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "spec-validator"
layer = "Core"
cli_tool = "claude"
persona = "carthos"
terraphim_role = "Carthos Architecture"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Validate specifications and design documents against implementation.
Ensure requirements traceability and design consistency.

## Domain Knowledge Search
When you need architecture context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Carthos Architecture"

## MANDATORY: Post verdict to Gitea
Post your spec validation verdict to the relevant Gitea issue.
- PASS if specs match implementation
- FAIL if spec violations found

Post verdict with @adf:merge-coordinator trigger.
Replace ISSUE_NUMBER with the issue from mention context, or the default tracking issue if no mention context.
"""
schedule = "0 4 * * *"
capabilities = ["specification-validation", "requirements-traceability", "design-review"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "specification requirement invariant domain model" --role "Carthos Architecture" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "compliance-watchdog"
layer = "Core"
cli_tool = "claude"
persona = "vigil"
terraphim_role = "Vigil Security"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Monitor compliance with security policies, coding standards, and
regulatory requirements. Flag violations and suggest remediation.

## Domain Knowledge Search
When you need security/compliance context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Vigil Security"

## MANDATORY: Post verdict to Gitea
Post your compliance verdict to the relevant Gitea issue.
- PASS if no compliance issues found
- FAIL if compliance violations found

Post verdict with @adf:merge-coordinator trigger.
Replace ISSUE_NUMBER with the issue from mention context, or the default tracking issue if no mention context.
"""
schedule = "0 5 * * *"
capabilities = ["compliance", "policy-enforcement", "standards-checking"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "compliance policy secrets credential authentication" --role "Vigil Security" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "upstream-synchronizer"
layer = "Core"
cli_tool = "claude"
persona = "conduit"
terraphim_role = "Conduit DevOps"
skill_chain = ["disciplined-implementation", "disciplined-verification"]
task = """Sync with upstream repositories. Check for new releases of key
dependencies, update lockfiles, and verify compatibility.

## Domain Knowledge Search
When you need DevOps context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Conduit DevOps"
"""
schedule = "0 3 * * *"
capabilities = ["sync", "dependency-management", "upstream-tracking"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "deployment dependency container infrastructure" --role "Conduit DevOps" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "documentation-generator"
layer = "Core"
cli_tool = "claude"
persona = "mneme"
terraphim_role = "Mneme Knowledge"
skill_chain = ["disciplined-implementation", "disciplined-verification"]
task = """Generate and maintain documentation. Update API docs, README files,
and knowledge base entries based on recent code changes.

## Domain Knowledge Search
When you need knowledge management context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Mneme Knowledge"
"""
schedule = "0 6 * * *"
capabilities = ["documentation", "knowledge-management", "api-docs"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "documentation knowledge management learning pattern" --role "Mneme Knowledge" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

# =============================================================================
# GROWTH LAYER (on-demand / scheduled)
# =============================================================================

[[agents]]
name = "compound-review"
layer = "Growth"
cli_tool = "claude"
persona = "carthos"
terraphim_role = "Carthos Architecture"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Perform compound reviews of recent changes across the codebase.
Identify cross-cutting concerns, architectural drift, and integration issues.

## Domain Knowledge Search
When you need architecture context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Carthos Architecture"
"""
schedule = "0 2 * * *"
capabilities = ["code-review", "architecture-review", "cross-cutting-analysis"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "architecture review quality attribute trade-off" --role "Carthos Architecture" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "quality-coordinator"
layer = "Growth"
cli_tool = "claude"
persona = "carthos"
terraphim_role = "Carthos Architecture"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Run compound code review for the issue referenced in Mention Context.
Post verdict as "compound-review verdict: GO" or "compound-review verdict: NO-GO".

If GO, trigger the 4 remaining reviewers via @adf: mentions:
@adf:test-guardian, @adf:security-sentinel, @adf:spec-validator, @adf:compliance-watchdog.

Replace ISSUE_NUMBER with the issue from mention context.
"""
capabilities = ["code-review", "quality-gate", "review-coordination"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "code review quality gate architecture" --role "Carthos Architecture" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "browser-qa"
layer = "Growth"
cli_tool = "claude"
persona = "lux"
terraphim_role = "Lux Frontend"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Test frontend features in the browser. Verify accessibility,
responsive design, and visual consistency across viewports.

## Domain Knowledge Search
When you need frontend context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Lux Frontend"
"""
capabilities = ["browser-testing", "accessibility-audit", "visual-qa"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "accessibility component responsive visual Svelte" --role "Lux Frontend" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "merge-coordinator"
layer = "Growth"
cli_tool = "claude"
persona = "conduit"
terraphim_role = "Conduit DevOps"
skill_chain = ["disciplined-research", "disciplined-verification"]
task = """Evaluate whether all 5 required code reviews have been posted on a Gitea issue,
and if all reviews are positive, merge the linked PR.

Required reviewers: compound-review, test-guardian, spec-validator, compliance-watchdog, security-sentinel.
Read all comments, parse verdicts (GO/PASS = positive, NO-GO/FAIL = negative).
If all 5 passed: merge PR and close issue.
If any missing or failed: post status summary and wait.

Replace ISSUE_NUMBER with the issue from mention context.
"""
capabilities = ["merge-coordination", "ci-cd", "conflict-resolution"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "deployment pipeline CI merge release" --role "Conduit DevOps" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "market-analyst"
layer = "Growth"
cli_tool = "claude"
persona = "meridian"
terraphim_role = "Meridian Research"
skill_chain = ["disciplined-research"]
task = """Research market trends, competitive landscape, and user needs.
Produce insights for product roadmap decisions.

## Domain Knowledge Search
When you need market research context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Meridian Research"
"""
schedule = "0 7 * * *"
capabilities = ["market-research", "competitive-analysis", "user-research"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "market research product strategy competitive analysis" --role "Meridian Research" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

[[agents]]
name = "knowledge-curator"
layer = "Growth"
cli_tool = "claude"
persona = "mneme"
terraphim_role = "Mneme Knowledge"
skill_chain = ["disciplined-research"]
task = """Curate and improve the knowledge graph. Identify gaps in domain
terminology, add new synonyms, and maintain KG quality.

## Domain Knowledge Search
When you need knowledge management context, use:
  terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "<query>" --role "Mneme Knowledge"
"""
schedule = "0 8 * * *"
capabilities = ["knowledge-curation", "kg-maintenance", "terminology-management"]

[agents.pre_check]
kind = "shell"
script = """terraphim-agent --config /opt/ai-dark-factory/persona_roles_config.json search "knowledge graph learning pattern decision framework" --role "Mneme Knowledge" --limit 5 2>/dev/null || echo '[]'"""
timeout_secs = 30

# =============================================================================
# FLOW DAG -- Compound Review v2
# =============================================================================

[[flows]]
name = "compound-review-v2"
schedule = "0 2 * * *"
repo_path = "/opt/ai-dark-factory/workspace"
base_branch = "main"
timeout_secs = 7200

[[flows.steps]]
name = "gather-changes"
kind = "action"
command = "git diff main..HEAD --name-only > /tmp/changed_files.txt && cat /tmp/changed_files.txt"
timeout_secs = 300
on_fail = "abort"

[[flows.steps]]
name = "analyze-architecture"
kind = "agent"
cli_tool = "claude"
model = "sonnet"
task = "Review the architecture changes in the modified files. Focus on design patterns, API contracts, and system boundaries."
timeout_secs = 900
on_fail = "skip_failed"

[[flows.steps]]
name = "check-security"
kind = "agent"
cli_tool = "claude"
model = "sonnet"
task = "Perform a security audit on the changed code. Look for injection vulnerabilities, unsafe deserialization, and authentication bypasses."
timeout_secs = 600
on_fail = "continue"

[[flows.steps]]
name = "gate-approval"
kind = "gate"
condition = "{{steps.analyze-architecture.exit_code}} == 0"

[[flows.steps]]
name = "checkpoint-completed"
kind = "checkpoint"