aiconnai

Use for an Engram dependency triage loop that watches advisories and lockfiles, proposes patch-only updates, rejects major upgrades without approval, and records reproducible verification commands.

Use for an Engram CI sweeper triage loop that groups CI failures, classifies root causes, caps attempts, and proposes bounded minimal-fix candidates only after budget, denylist, and verifier gates.

Use for an Engram daily triage loop that safely collects and classifies CI, issue, PR, commit, and chat signals into report-only handoff state with bounded scope, no auto-fix, and loop-run-log updates.

Use for an Engram PR babysitter triage loop that monitors PR aging, CI blocks, reviewer threads, and lightweight action proposals with isolated-worktree fix candidates and human escalation for risky changes.

Use when designing, implementing, auditing, hardening, or operating agent loops in this repository, including Claude Code/Grok Build/Codex style `/loop` workflows, explicit `/goal` runs, Git worktree orchestration, MCP and memory automation, sub-agent execution, STATE.md handoffs, cost controls, maker/checker verification, and the Engram security gate (full `sensors.sh` + AgentShield scan + `just ci`). Loops default to L1 report-only; autonomous execution against Engram requires a reviewed ADR per the harness security contract.

Weekly or manual security triage loop that executes AgentShield with a hard iteration cap and records the result in a repository state file.

Run structured consensus with Engram's `memory_council` tool and `CouncilSkill` SDK wrappers for architecture, design, policy, security, reliability, performance, ADR, or tradeoff decisions. Use when the user asks for consensus, council review, multi-agent-style review, decision rationale, option comparison, or mentions `llm-council`, `memory_council`, or `CouncilSkill`. Do not use for simple lookup, single-answer retrieval, or execution-only coding tasks after a decision is already made.

Bootstrap an AI agent's engram memory session in a new repository. Guides through loading prior context, project scan, workspace setup, first memory creation, search, and session close. Use when starting work in a repo that has engram configured, when setting up memory for the first time, or when a new agent needs to establish context. Skip if the session is already active and workspaces are already configured.

Use for a Codex dependency sweeper triage loop that watches advisories and lockfiles, proposes patch-only updates, rejects major upgrades without approval, and records reproducible verification commands.

Use when designing, implementing, auditing, hardening, or operating autonomous agent loops in this repository, including Codex/Grok/Claude Code style `/loop` workflows, explicit `/goal` runs, Git worktree orchestration, connector-backed automation, sub-agent execution, STATE.md handoffs, cost controls, maker/checker verification, and AgentShield security gates.

Use for a Codex CI Sweeper triage loop that groups CI failures, classifies root causes, caps attempts, and proposes bounded minimal-fix candidates only after budget, denylist, and verifier gates.

Use for a Codex daily triage loop that safely collects and classifies CI, issue, PR, commit, and chat signals into report-only handoff state with bounded scope, no auto-fix, and loop-run-log updates.

Use for a Codex PR Babysitter triage loop that monitors PR aging, CI blocks, reviewer threads, and lightweight action proposals with isolated-worktree fix candidates and human escalation for risky changes.

Use when working with Huly workspaces, projects, issues, labels, components, milestones, or documents. Prefer the official Huly Platform API for token-based automation, especially when the user authenticates to Huly through GitHub/Google/OIDC and the third-party huly-cli email/password login path cannot be used.

Demo Hermes skill fixture.

Decomposes complex tasks into a DAG of atomic subtasks with model routing and dependency mapping. Use when the user wants to plan a multi-file feature, refactor, or system build without executing it yet. Triggers on "plan", "break down", "decompose", "task graph", or when the user wants to understand the work before committing to execution.

Performs structured code review across five dimensions: maintainability, security, performance, design, and completeness. Fixes issues before presenting code to the user — the review loop runs automatically, making every output a second draft, not a first. Triggers on "review", "code review", "check my code", or automatically after implementation tasks complete.

Enforces strict test-driven development via a RED-GREEN-REFACTOR state machine with a context whiteboard and deterministic validation gates. Use before writing any implementation code — write the failing test first, watch it fail, then implement. Triggers on "implement", "build", "fix", "add feature", or any task that produces code changes. Also use when the user says "test first", "TDD", or "red green refactor".

Sets up and manages isolated git worktrees for parallel agent work. Use when you need to run multiple agents on independent tasks simultaneously, or when any work benefits from branch isolation. Triggers on "worktree", "parallel", "isolate", or when dispatching concurrent agents.

Orchestrates multi-agent workflows with intelligent model tiering (Opus/Sonnet/Haiku), git worktrees for parallelism, DAG-based task scheduling, and self-correction loops. Use this skill whenever the user asks to implement a complex feature, build a system, refactor a large codebase, or any task that benefits from decomposition into parallel subtasks. Also trigger when the user mentions "multi-agent", "orchestrate", "parallelize", "decompose", "plan and execute", or wants to optimize Claude Code usage costs. Even if the user just says "implement X" where X is clearly multi-file or multi-concern, use this skill to decompose and coordinate efficiently.

Autonomous penetration testing via Shannon — a white-box security testing framework that executes real attacks and reports only confirmed exploits. Pairs with mao-security (static analysis) to provide full security coverage: static review on every PR, dynamic pentest before major releases. Requires Docker and Shannon installed: npx skills add unicodeveloper/shannon