The library

os-eval-lab-setup · Bootstrapping isolated test repos with consistent eval harness before autoresearch iterations.

gsd:resume-at · Scheduling future Claude sessions around usage caps or scheduled overnight batch work.

os-eval-runner · Autonomously iterating skill improvements using empirical scoring gates instead of human feedback.

gsd:resume-work · Seamlessly restoring GSD session state from checkpoints across multiple Claude conversations.

os-evolution-planner · Proposing and executing agent evolution paths (update/create/orchestrate) with structured planning.

gsd:review-backlog · Systematically triaging project tasks and surfacing critical path blockers.

os-evolution-verifier · Validating that evolved agents meet their original requirements before applying to production.

gsd:review · Gate-checking workflow completion before advancing to next phase.

os-experiment-log · Durable tracking of agent experiments with structured metadata and cross-run analysis.

gsd:scan · Quick status checkup showing active phase, pending tasks, and system health.

clinical-nlp-extractor · Extracting structured medical data from unstructured clinical notes at scale.

os-guide · Onboarding users to Agentic OS workflows without requiring manual documentation reading.

gsd:secure-phase · Retroactively auditing threat mitigations before shipping a production phase

os-improvement-loop · Multi-agent systems needing concurrent eval-driven improvement cycles

gsd:session-report · Generating shareable post-session summaries with estimated token costs

os-improvement-report · Tracking and visualizing agent loop improvement trends over cycles

gsd:set-profile · Switching GSD agent quality/cost tradeoffs without restarting work

os-init · Bootstrapping Agentic OS structure for new multi-agent projects

gsd:settings · Interactive configuration of GSD workflow toggles and model profiles

os-memory-manager · Closing sessions with proper three-tier memory promotion and deduplication

gsd:ship · Creating PRs and tracking merges after verify-work passes

self-evolution · Self-healing failed selectors and stale scripts via tier-gated edits

gsd:sketch-wrap-up · Distilling throwaway sketches into persistent design skills

todo-check · Pre-flight auditing a file for technical debt before committing

gsd:sketch · Exploring design directions with fast HTML mockup variants

agent-swarm · Partitioning large features into parallel independent sub-tasks

gsd:spec-phase · Clarifying phase requirements with Socratic interview before discuss-phase

dual-loop · Balancing dual concurrent execution cycles with separate eval gates

gsd:spike-wrap-up · Converting research findings into reusable project-local skills

learning-loop · Continuous improvement of agent skills through feedback loops

gsd:spike · Fast exploratory research on technical unknowns before planning

orchestrator · Coordinating multi-agent task dispatch across a distributed team

gsd:stats · When you need a single authoritative view of all project phases, plans, and metrics across the entire codebase.

red-team-review · When complex work needs independent adversarial critique before commitment, especially for architectures and research.

gsd:thread · When work spans multiple sessions but doesn't belong to a specific phase and needs persistent context.

triple-loop-learning · When you want autonomous continuous improvement of system instructions based on objective headless benchmarks.

gsd:ui-phase · When starting a frontend phase and need a design specification contract from research to verification.

memory-management · When you need a tiered recall system balancing hot-cache speed with deep semantic search across sessions.

gitnexus-exploring · When learning a large unfamiliar codebase and need execution flows and symbol relationships mapped.

rlm-audit · When you've added new files and need to identify exactly what's missing from the semantic cache.

rlm-cleanup-agent · When files are deleted or moved and the RLM ledger contains stale entries to remove.

gsd:undo · When you need to safely revert phase or plan commits with dependency checking, not blind git reset.

rlm-curator · When you need to distill files into the RLM Summary Ledger and keep the cache in sync with codebase.

rlm-distill-agent · When you have uncached files and need high-quality 1-sentence summaries written to the ledger.

gsd:validate-phase · When a phase is complete but validation coverage is unknown and needs retroactive audit and gap-filling.

rlm-init · When setting up RLM for the first time in a project or adding a new cache profile.

gsd:verify-work · When you want conversational user acceptance testing with automatic diagnosis of found issues.

rlm-search · When searching for architecture, code, or documentation and need O(1) RLM scan before semantic or exact search.

gsd:workstreams · When validating project work and need to execute gsd:validate-phase workflow.

vector-db-audit · When you need O(1) keyword lookup across RLM summaries without semantic search overhead.