Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Latest commit

 

History

42 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Catalog

This repository is a growing catalog of reusable AI agent components organized by project. It is workflow-first: small, focused agents are composed into explicit pipelines.

Goals

  • Keep each agent self-contained and easy to run.
  • Reuse project-level patterns so new agents are quick to add.
  • Document every agent clearly as the catalog scales.

Repository Structure

  • AGENTS.md - contribution and catalog conventions (slim); full agent/project index lives in catalog/README.md.
  • .agents/ - Agent Skills (project scope); see .agents/README.md (skills under .agents/skills/, per Where to scan).
  • catalog/ - all projects and their agents; open catalog/README.md for the canonical catalog index.
  • docs/ - cross-cutting architecture and standards.
catalog/
  projects/
    <project>/
      README.md
      agents/
        <agent>/
          agent.yaml
          prompts/
          workflows/
          examples/
          tests/

Current Projects

  1. starter-kit
    • heartbeat-agent: simple status summarizer and health signal formatter.
    • classifier-agent: classifies text into one intent label with confidence.
  2. support-ops
    • triage-agent: converts inbound support text into priority/category/next-action.
    • reply-drafter-agent: drafts concise customer reply subject/body from structured triage fields.
    • summary-agent: summarizes weekly support ticket trends and follow-up actions.
    • handoff-agent: generates shift-transition briefs from active incidents.
  3. security-ops
    • agentic-security-scanner-agent: scans repo controls and maps findings to OWASP ASI categories.
  4. planner-executor
    • planner-agent: converts a goal and constraints into a bounded execution plan.
    • executor-agent: summarizes progress and completion state from plan steps.
  5. research-ops
    • source-planner-agent: plans what evidence to collect next from a research question.
    • retrieval-agent: extracts bounded notes from query + source hints.
    • gap-detector-agent: detects unsupported assertions and collection gaps.
    • synthesis-agent: converts notes into audience-aware summary and actions.
  6. knowledge-ops
    • evidence-ranker-agent: scores and ranks candidate evidence for downstream use.
    • claim-trace-agent: maps assertions to evidence references and support states.
    • memory-curator-agent: distills run artifacts into reusable memory entries.
    • temporal-watch-agent: compares snapshots over time and emits drift signals.
  7. qa-ops
    • test-case-generator-agent: generates bounded QA scenarios from feature requirements.
    • regression-triage-agent: classifies regression cause/severity and proposes follow-up actions.
  8. workflow-ops
    • router-agent: routes work items to the best-fit agent with a priority.
    • dependency-router-agent: routes work only when prerequisites are satisfied.
    • retry-policy-agent: decides retry/backoff/escalation policy for failed stages.
    • checkpoint-agent: records workflow stage/status checkpoints for traceability.
  9. control-ops
    • lineage-recorder-agent: structures decision events into append-only lineage records.
    • scope-validator-agent: validates proposed actions against governance requirements with pass/review/fail gating.
    • exception-policy-agent: evaluates controlled policy exceptions with explicit owner and expiry.
    • approval-memory-agent: tracks approval state and expiration for governance recall.
    • blast-radius-assessor-agent: estimates blast radius from weighted permission, dependency, and resource-limit factors.
    • kill-path-auditor-agent: audits shutdown capabilities against the four-level kill path spectrum and optional ISO last_tested recency.
  10. data-ops
  • schema-drift-detector-agent: detects schema changes between versions and classifies drift severity.
  • data-validator-agent: validates data records against rules and reports violations.
  1. code-ops
    • code-reviewer-agent: reviews code diffs for security, correctness, and style issues.
    • pr-summary-agent: summarizes PR changes for reviewers with risk assessment.
  2. observability-ops
    • log-analyzer-agent: analyzes log entries for patterns and anomalies.
    • slo-reporter-agent: generates SLO compliance reports from service metrics and targets.
    • change-correlation-agent: correlates incident signal shifts with nearby deploy/config events.
    • alert-tuner-agent: suggests alert threshold tuning from historical noise patterns.
  3. eval-ops
    • benchmark-curator-agent: deduplicates eval cases and surfaces benchmark coverage gaps.
    • regression-score-agent: compares baseline vs current eval metrics for regressions.
    • quality-drift-reporter-agent: summarizes quality trend and drift across time windows.
  4. experiment-ops
    • hypothesis-registration-agent: registers and normalizes hypotheses with clarity checks.
    • experiment-plan-agent: generates bounded experiment plans from a hypothesis.
    • result-adjudication-agent: maps observed metrics to supports, inconclusive, or refutes verdicts.
  5. artifact-ops
    • artifact-inventory-agent: normalizes run artifacts with roles and content digests for bundles.
    • bundle-manifest-agent: builds ordered manifest entries and an aggregate bundle root hash.
    • bundle-seal-agent: emits bundle id, seal status, lineage attachment, and verification steps.
  6. failure-ops
    • failure-library-agent: normalizes incident observations into bounded failure mode records.
    • blast-pattern-cluster-agent: clusters failure modes into blast-pattern groups.
    • rollback-playbook-agent: generates rollback playbooks with prerequisites, abort conditions, and verification checks.
  7. inter-ops
    • schema-compat-validator-agent: validates producer/consumer schema compatibility from payload snapshots.
  8. cost-ops
    • cost-attribution-agent: attributes token/runtime spend by stage.
    • budget-guardrail-agent: evaluates attributed spend against budget limits.
    • pipeline-optimizer-agent: emits ranked optimization actions for pipeline cost.
  9. agent-incident-drill
    • Scenario project that composes existing catalog agents into a measurable incident-response drill with governance, lineage, blast-radius, kill-path, rollback, and scorecard artifacts.

Run Agents Locally

  • python3 scripts/run_agent.py --agent starter-kit.heartbeat-agent --input catalog/projects/starter-kit/agents/heartbeat-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent starter-kit.classifier-agent --input catalog/projects/starter-kit/agents/classifier-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent support-ops.triage-agent --input catalog/projects/support-ops/agents/triage-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent support-ops.reply-drafter-agent --input catalog/projects/support-ops/agents/reply-drafter-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent support-ops.summary-agent --input catalog/projects/support-ops/agents/summary-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent support-ops.handoff-agent --input catalog/projects/support-ops/agents/handoff-agent/examples/example-input.json --pretty
  • python3 scripts/run_support_pipeline.py --input catalog/projects/support-ops/examples/pipeline-input.json --pretty
  • python3 scripts/run_agent.py --agent planner-executor.planner-agent --input catalog/projects/planner-executor/agents/planner-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent planner-executor.executor-agent --input catalog/projects/planner-executor/agents/executor-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent research-ops.source-planner-agent --input catalog/projects/research-ops/agents/source-planner-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent research-ops.retrieval-agent --input catalog/projects/research-ops/agents/retrieval-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent research-ops.gap-detector-agent --input catalog/projects/research-ops/agents/gap-detector-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent research-ops.synthesis-agent --input catalog/projects/research-ops/agents/synthesis-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent knowledge-ops.evidence-ranker-agent --input catalog/projects/knowledge-ops/agents/evidence-ranker-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent knowledge-ops.claim-trace-agent --input catalog/projects/knowledge-ops/agents/claim-trace-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent knowledge-ops.memory-curator-agent --input catalog/projects/knowledge-ops/agents/memory-curator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent knowledge-ops.temporal-watch-agent --input catalog/projects/knowledge-ops/agents/temporal-watch-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent qa-ops.test-case-generator-agent --input catalog/projects/qa-ops/agents/test-case-generator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent qa-ops.regression-triage-agent --input catalog/projects/qa-ops/agents/regression-triage-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent eval-ops.benchmark-curator-agent --input catalog/projects/eval-ops/agents/benchmark-curator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent eval-ops.regression-score-agent --input catalog/projects/eval-ops/agents/regression-score-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent eval-ops.quality-drift-reporter-agent --input catalog/projects/eval-ops/agents/quality-drift-reporter-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent experiment-ops.hypothesis-registration-agent --input catalog/projects/experiment-ops/agents/hypothesis-registration-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent experiment-ops.experiment-plan-agent --input catalog/projects/experiment-ops/agents/experiment-plan-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent experiment-ops.result-adjudication-agent --input catalog/projects/experiment-ops/agents/result-adjudication-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent artifact-ops.artifact-inventory-agent --input catalog/projects/artifact-ops/agents/artifact-inventory-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent artifact-ops.bundle-manifest-agent --input catalog/projects/artifact-ops/agents/bundle-manifest-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent artifact-ops.bundle-seal-agent --input catalog/projects/artifact-ops/agents/bundle-seal-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent failure-ops.failure-library-agent --input catalog/projects/failure-ops/agents/failure-library-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent failure-ops.blast-pattern-cluster-agent --input catalog/projects/failure-ops/agents/blast-pattern-cluster-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent failure-ops.rollback-playbook-agent --input catalog/projects/failure-ops/agents/rollback-playbook-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent inter-ops.schema-compat-validator-agent --input catalog/projects/inter-ops/agents/schema-compat-validator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent cost-ops.cost-attribution-agent --input catalog/projects/cost-ops/agents/cost-attribution-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent cost-ops.budget-guardrail-agent --input catalog/projects/cost-ops/agents/budget-guardrail-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent cost-ops.pipeline-optimizer-agent --input catalog/projects/cost-ops/agents/pipeline-optimizer-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent workflow-ops.router-agent --input catalog/projects/workflow-ops/agents/router-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent workflow-ops.dependency-router-agent --input catalog/projects/workflow-ops/agents/dependency-router-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent workflow-ops.retry-policy-agent --input catalog/projects/workflow-ops/agents/retry-policy-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent workflow-ops.checkpoint-agent --input catalog/projects/workflow-ops/agents/checkpoint-agent/examples/example-input.json --pretty
  • python3 scripts/run_planner_executor_pipeline.py --input catalog/projects/planner-executor/examples/pipeline-input.json --pretty
  • python3 scripts/run_workflow_pipeline.py --input catalog/projects/workflow-ops/examples/pipeline-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.lineage-recorder-agent --input catalog/projects/control-ops/agents/lineage-recorder-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.scope-validator-agent --input catalog/projects/control-ops/agents/scope-validator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.exception-policy-agent --input catalog/projects/control-ops/agents/exception-policy-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.approval-memory-agent --input catalog/projects/control-ops/agents/approval-memory-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.blast-radius-assessor-agent --input catalog/projects/control-ops/agents/blast-radius-assessor-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent control-ops.kill-path-auditor-agent --input catalog/projects/control-ops/agents/kill-path-auditor-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent data-ops.schema-drift-detector-agent --input catalog/projects/data-ops/agents/schema-drift-detector-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent data-ops.data-validator-agent --input catalog/projects/data-ops/agents/data-validator-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent code-ops.code-reviewer-agent --input catalog/projects/code-ops/agents/code-reviewer-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent code-ops.pr-summary-agent --input catalog/projects/code-ops/agents/pr-summary-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent observability-ops.log-analyzer-agent --input catalog/projects/observability-ops/agents/log-analyzer-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent observability-ops.slo-reporter-agent --input catalog/projects/observability-ops/agents/slo-reporter-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent observability-ops.change-correlation-agent --input catalog/projects/observability-ops/agents/change-correlation-agent/examples/example-input.json --pretty
  • python3 scripts/run_agent.py --agent observability-ops.alert-tuner-agent --input catalog/projects/observability-ops/agents/alert-tuner-agent/examples/example-input.json --pretty
  • python3 scripts/run_governance_pipeline.py --input catalog/projects/control-ops/examples/governance-pipeline-input.json --pretty
  • python3 scripts/run_resilience_pipeline.py --input catalog/projects/control-ops/examples/resilience-pipeline-input.json --pretty
  • python3 scripts/run_incident_pipeline.py --input examples/incident-pipeline-input.json --pretty
  • python3 scripts/run_agent_incident_drill.py --input catalog/projects/agent-incident-drill/examples/drill-input.json --pretty
  • make compare-agent-incident-drill-scorecards-example (diff scorecards from two drill runs)
  • python3 scripts/run_security_scan.py --target-path . --pretty
  • python3 scripts/run_security_scan.py --target-path /path/to/other/catalog --rules custom-rules.json --pretty
  • AGENT_MODE=llm python3 scripts/run_support_pipeline.py --input catalog/projects/support-ops/examples/pipeline-input.json --pretty
  • make state-up to start Redis for pipeline state persistence
  • make llm-up && make llm-pull for a speed-first local model (llama3.2:3b)

See /Users/macos-user/.projects/stack-research/agents/docs/local-usage.md for full usage.

End-to-End Story: Incident Day Walkthrough

A new engineer joins the on-call rotation. Mid-morning, support reports that some customers cannot log in after a release. Instead of jumping between tools and ad-hoc notes, the engineer uses the catalog as a structured agentic workflow.

The engineer starts by routing the task, then triaging the issue, generating QA scenarios, triaging an observed regression failure, producing a stakeholder-ready summary, and recording a checkpoint. The point is not replacing engineering judgment. The point is making the system legible, repeatable, and fast under pressure.

1) Route the incoming task

python3 scripts/run_agent.py \
  --agent workflow-ops.router-agent \
  --input catalog/projects/workflow-ops/agents/router-agent/examples/example-input.json \
  --pretty

Example output:

{
  "priority": "p2",
  "rationale": "Support issue intent detected; route to triage.",
  "target_agent": "support-ops.triage-agent"
}

2) Triage the support issue

python3 scripts/run_agent.py \
  --agent support-ops.triage-agent \
  --input catalog/projects/support-ops/agents/triage-agent/examples/example-input.json \
  --pretty

Example output:

{
  "category": "access",
  "next_action": "Escalate to auth on-call and collect user and timestamp details.",
  "priority": "p2"
}

3) Generate QA coverage for the suspected area

python3 scripts/run_agent.py \
  --agent qa-ops.test-case-generator-agent \
  --input catalog/projects/qa-ops/agents/test-case-generator-agent/examples/example-input.json \
  --pretty

Example output:

{
  "risk_focus": "medium",
  "test_cases": [
    "Happy path: validate SSO login flow with Users can sign in with SAML",
    "Validation edge: reject invalid input for SSO login flow",
    "Boundary check: enforce limits and defaults for SSO login flow",
    "Failure path: verify clear error handling for SSO login flow",
    "Security check: block unauthorized access during SSO login flow"
  ]
}

4) Triage a regression failure signal

python3 scripts/run_agent.py \
  --agent qa-ops.regression-triage-agent \
  --input catalog/projects/qa-ops/agents/regression-triage-agent/examples/example-input.json \
  --pretty

Example output:

{
  "probable_cause": "dependency",
  "recommended_actions": [
    "Reproduce failure with focused logs for: Production timeout after dependency version update",
    "Compare failure window with most recent merged changes",
    "Review changed components: auth-service, sdk-client"
  ],
  "severity": "sev2"
}

5) Synthesize findings for stakeholders

python3 scripts/run_agent.py \
  --agent research-ops.synthesis-agent \
  --input catalog/projects/research-ops/agents/synthesis-agent/examples/example-input.json \
  --pretty

Example output:

{
  "headline": "Security Brief Summary",
  "next_actions": [
    "Validate highest-impact claim with one primary source",
    "Document assumptions and unresolved risks",
    "Share summary with stakeholders for review"
  ],
  "summary": "Key findings: Research objective: Summarize ASI09 mitigation guidance; Source note: enforce output contracts; Source note: require explicit human approval for sensitive actions"
}

6) Record a checkpoint in the workflow

python3 scripts/run_agent.py \
  --agent workflow-ops.checkpoint-agent \
  --input catalog/projects/workflow-ops/agents/checkpoint-agent/examples/example-input.json \
  --pretty

Example output:

{
  "checkpoint_id": "release-2026-02-16:qa-validation:in_progress",
  "recorded": true,
  "summary": "Checkpoint recorded for workflow release-2026-02-16 at stage qa-validation with status in_progress. Notes: Integration tests running on staging."
}

This is the core advantage of a good agentic system for engineering teams: clear contracts, composable steps, and traceable state across the full lifecycle of work.

Workflow-Ops Pipeline Example

For a fast orchestration-only path, run the composed workflow pipeline directly:

python3 scripts/run_workflow_pipeline.py \
  --input catalog/projects/workflow-ops/examples/pipeline-input.json \
  --pretty

This pipeline composes:

  1. workflow-ops.router-agent
  2. the routed target agent
  3. workflow-ops.checkpoint-agent

and returns a single structured object with route decision, target output, checkpoint record, and pipeline_status.

Governance Pipeline (run_governance_pipeline.py)

scope-validator -> [target agent] -> lineage-recorder -> checkpoint
  • Validates action scope/permissions/reversibility before executing the target agent
  • If scope validation returns fail, the pipeline short-circuits: records lineage ("blocked by governance gate") and a failed checkpoint, but never runs the target
  • If pass or review, proceeds to execute the target, then records full decision lineage and a completion checkpoint
  • Returns pipeline_status: ok, blocked, or degraded
python3 scripts/run_governance_pipeline.py \
  --input catalog/projects/control-ops/examples/governance-pipeline-input.json \
  --pretty

Resilience Pipeline (run_resilience_pipeline.py)

blast-radius-assessor -> kill-path-auditor
  • Assesses blast radius first (risk score, damage potential, detection/containment speeds)
  • Then audits kill path coverage against the same system
  • Computes a combined resilience_verdict:
    • adequate: full kill path coverage (4/4)
    • partial: moderate coverage
    • at-risk: high risk score + low coverage
    • inadequate: high risk score + very low coverage

Both follow the standard degraded-mode fallback pattern on validation failures. 10 new tests cover happy path, short-circuit blocking, degraded modes, and verdict logic.

python3 scripts/run_resilience_pipeline.py \
  --input catalog/projects/control-ops/examples/resilience-pipeline-input.json \
  --pretty

Incident Pipeline (run_incident_pipeline.py)

router -> triage -> test-case-generator -> synthesis -> scope-validator -> checkpoint

Cross-domain orchestration that chains 6 agents from 5 different domains into a single incident response flow:

  1. Route (workflow-ops): classifies the incident and selects a target agent
  2. Triage (support-ops): assigns priority, category, and next action
  3. QA (qa-ops): generates test cases for the affected feature area
  4. Synthesis (research-ops): combines findings into a stakeholder summary
  5. Governance (control-ops): validates the proposed response action against scope/permissions and stops on review or fail
  6. Checkpoint (workflow-ops): records the pipeline outcome for traceability

If governance returns verdict: fail, the pipeline status is blocked. Validation failures at any stage produce degraded status with prior stage outputs preserved.

python3 scripts/run_incident_pipeline.py \
  --input examples/incident-pipeline-input.json \
  --pretty

Pipeline State Persistence (Redis)

Pipelines can optionally persist intermediate stage outputs and final results to Redis. This enables multi-turn workflows, debugging, and auditing.

make state-up                    # start Redis
make state-down                  # stop Redis

Add --state to any pipeline runner to enable persistence:

python3 scripts/run_support_pipeline.py \
  --input catalog/projects/support-ops/examples/pipeline-input.json \
  --state --pretty

Optionally provide --run-id to set a custom run identifier; otherwise one is auto-generated.

State is stored with a 1-hour TTL and auto-expires. If Redis is unavailable, pipelines work exactly as before (stateless, single-shot).

Test Suite

  • python3 -m unittest discover -s tests -v
  • make test
  • make verify-env (policy check + deterministic tests, with deterministic security suite in non-dev environments)
  • make test-security (ASI01+ASI02+ASI03+ASI04+ASI05+ASI06+ASI07+ASI08+ASI09+ASI10 adversarial checks + scanner tests)
  • make test-security-llm (ASI01+ASI02+ASI03+ASI04+ASI05+ASI06+ASI07+ASI08+ASI09+ASI10 adversarial checks against local LLM)
  • make test-integration-llm (optional LLM-backed checks)

The test suite currently includes:

  • data-ops deterministic behavior tests (schema-drift-detector/data-validator).
  • code-ops deterministic behavior tests (code-reviewer/pr-summary).
  • observability-ops deterministic and LLM behavior tests (log-analyzer/slo-reporter/change-correlation/alert-tuner).
  • knowledge-ops deterministic and LLM behavior tests (evidence-ranker/claim-trace/memory-curator/temporal-watch).
  • control-ops deterministic and LLM behavior tests (lineage-recorder/scope-validator/exception-policy/approval-memory/blast-radius-assessor/kill-path-auditor).
  • failure-ops deterministic and LLM behavior tests (failure-library/blast-pattern-cluster/rollback-playbook).
  • inter-ops deterministic and LLM behavior tests (schema-compat-validator).
  • cost-ops deterministic and LLM behavior tests (cost-attribution/budget-guardrail/pipeline-optimizer).
  • governance pipeline composition tests.
  • resilience pipeline composition tests.
  • behavior tests for local agent runtime logic.
  • support-ops deterministic and LLM behavior tests (triage/reply/summary/handoff).
  • pipeline composition tests.
  • ASI01 goal-hijack adversarial regression tests.
  • ASI01 goal-hijack LLM adversarial regression tests.
  • ASI02 tool-misuse adversarial regression tests.
  • ASI02 tool-misuse LLM adversarial regression tests.
  • ASI03 identity/privilege-abuse adversarial regression tests.
  • ASI03 identity/privilege-abuse LLM adversarial regression tests.
  • ASI04 supply-chain/runtime-source adversarial regression tests.
  • ASI04 supply-chain/runtime-source LLM adversarial regression tests.
  • ASI05 unexpected-code-execution adversarial regression tests.
  • ASI05 unexpected-code-execution LLM adversarial regression tests.
  • ASI06 memory/context-poisoning adversarial regression tests.
  • ASI06 memory/context-poisoning LLM adversarial regression tests.
  • ASI07 inter-agent-communication adversarial regression tests.
  • ASI07 inter-agent-communication LLM adversarial regression tests.
  • ASI08 cascading-failure adversarial regression tests.
  • ASI08 cascading-failure LLM adversarial regression tests.
  • ASI09 human-agent-trust-exploitation adversarial regression tests.
  • ASI09 human-agent-trust-exploitation LLM adversarial regression tests.
  • ASI10 rogue-agent adversarial regression tests.
  • ASI10 rogue-agent LLM adversarial regression tests.
  • security scanner tests (glob discovery, custom rules, check operators, nested catalogs).
  • state store unit tests (NoOp fallback, pipeline helpers, live Redis integration).
  • agent.yaml JSON Schema validation (schemas/agent.json).
  • benchmark/eval fixtures: 86 cases across 24 agents (evals/cases.json per agent).
  • incident pipeline cross-domain composition tests.
  • catalog structure checks for required per-agent files.
  • optional integration tests against local Ollama.

Policy Pack

Environment ASI control baselines live in:

  • policy/asi-control-baselines.json

This policy pack defines required controls for ASI01 through ASI10 across:

  • dev
  • staging
  • prod

Reference documentation:

  • docs/policy-pack.md
  • python3 scripts/check_policy_pack.py --env dev --mode deterministic
  • make check-policy-pack and make check-policy-pack-llm

LLM-oriented make targets are policy-gated by environment (POLICY_ENV) and fail fast when policy disallows LLM mode.

Add a New Agent

  1. Create catalog/projects/<project>/agents/<agent-name>/.
  2. Add agent.yaml with purpose, IO contract, and runtime assumptions (validated by schemas/agent.json).
  3. Add prompt/workflow/example/test docs.
  4. Add evals/cases.json with benchmark fixtures (validated by schemas/eval-case.json).
  5. Update catalog/projects/<project>/README.md.
  6. Update root README.md, catalog/README.md (canonical catalog index), and AGENTS.md if conventions or Update Rule change.

Next Ideas

  1. CI pipeline (GitHub Actions for make test and make test-security).

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages