Loop engineering for AI coding agents — shaping the whole run, not one prompt: a Markdown-first overlay of task records, role boundaries, verification evidence, and review gates that holds across multi-hour work
Skill Claude CodeCodex
Use when an agent or the loop hits a wall it cannot clear on its own – provider outage, rate limit, missing credentials, an impossible or contradictory task, review deadlock, merge conflict, no implementation artifact produced, or an exhausted attempt budget / self-loop with no progress – and must record a durable…
Skill Claude CodeCodex
Use when a task is classified type: change-request because it changes a locked architecture, process, plan, or repository decision. Defines the docs-first approval gate that must pass before implementation and how files-backed or GitHub-backed task records are held blocked until then.
Skill Claude CodeCodex
Use the moment a required check, test, build, or runtime behavior fails or surprises – before proposing or applying any fix, including on every needsrevision iteration that involves a defect. Defines feedback-loop-first debugging, root-cause phases, calibrated hypotheses, tagged debug instrumentation, confirmed…
Skill Claude CodeCodex
Use when recording, updating, accepting, or superseding a durable project decision that constrains future work, including a maintainer's promotion of an already-recorded policy-level verification observation. Covers tracked Markdown decision records under .agenticloop/decisions/, maintainer ownership, proposed vs…
Skill Claude CodeCodex
Use when a role needs to emit an Agentic Loop workflow-gate event (role.invoked, task.created, task.started, check.run, review.started, review.result, decision.recorded, blocked, needscontext, task.closed, summary.published) and must resolve the event logging command, write concise event entries, or confirm that event…
Skill Claude CodeCodex
Use when implementing or revising any frontend UI – landing pages, marketing sites, portfolios, app screens, dashboards, or components – and when choosing layout, typography, color, spacing, motion, or styling. Enforces a design-read first, real design systems, anti-slop discipline, accessibility, and a pre-flight…
Skill Claude CodeCodex
Use whenever an agent authors or reads a GitHub issue, pull request, or comment body, or writes an agent-authored commit message. Covers the [[agent: ]] body trailer, Task:/Agent: commit trailer, and how to read cooperative attribution when multiple roles share one GitHub identity.
Skill Claude CodeCodex
Parked, human-invoked skill. Use when a human explicitly asks for a retrospective that turns repeated agent-process friction into durable improvements. Converts repeated blocked categories, check failures, review iterations, or recurring follow-up topics drawn from event logs into skill updates, docs entries, or…
Skill Claude CodeCodex
Use when an authorized multi-task work unit needs its required current Parallel Opportunity Scan, especially when 2 or more ready independent tasks may parallelize, or the orchestrator must decide serial versus parallel execution, or when planning, reviewing, joining, troubleshooting, or boundedly reconciling parallel…
Skill Claude CodeCodex
Use when the current role is maintainer or engineer and either the human explicitly asks for ponytail, lazy mode, YAGNI, simplest solution, minimal solution, do less, shortest path, or minimalist implementation discipline, or the active task record sets minimalism: lite, minimalism: full, or minimalism: ultra, or the…
Skill Claude CodeCodex
Use when the maintainer reviews an implementation artifact against its task record and decides accepted vs needsrevision, and when the engineer responds to review feedback. Defines GitHub review markers, files-backed reviewstatus, ordered three-lens review, evidence rules, disputed-items protocol, and mandatory triage…
Skill Claude CodeCodex
Use when the orchestrator starts or resumes a task loop, routes first-run project-map confirmation, creates a task record handoff, hands off implementation, routes review, or runs closeout. Ensures real host delegation happens when available, or an explicit bounded single-agent role assumption is used when it is not…
Skill Claude CodeCodex
Use when a maintainer or human is setting up Agentic Loop in a target project for the first time, when source document names are non-standard or ambiguous, when .agenticloop/project.md is still unconfirmed, or when a human wants to update its development-stage profile. Discovers bounded candidate project docs once…
Skill Claude CodeCodex
Use when preparing pre-certification closeout once covered tasks are accepted, running the final closeout gate once the resulting candidate is integrated or frozen and certified, finishing a human-identified work unit in a flat project, or cleaning up an accepted worktree after integration. Defines conditional…
Skill Claude CodeCodex
Use when creating, updating, or reading the task record for a single implementation task. Covers required task sections, no-placeholder rules, files-first task-file frontmatter, work-unit summary template, implementation artifact linkage, and safe GitHub projection posting practices. Backend projections are documented…
Skill Claude CodeCodex
Use when implementing any code change for a task – new behavior, a bugfix, or a revision after review – before writing the production code. Defines the failing-test-first cycle, public-seam testing, vertical slices, scaffold and infra handling, and counters to common excuses for tests-after-code.
Skill Claude CodeCodex
Use before claiming any work state – done, fixed, passing, green, complete, mergeable – and whenever a required or cited check times out, is unexpectedly expensive, or needs a retry. Defines the identify-run-read-verify gate, timeout evidence and retry records, and the baseline/lane-final/integrated/post-merge…
Skill Claude CodeCodex
Use when every task in a work unit is accepted and the whole unit must be certified by the auditor before closeout, when routing a non-certifying audit report into remediation tasks, or when a re-audit is required because remediation changed the audited artifact. Defines the audit packet, the six audit perspectives…