plurai-simulation

Plugin plurai-simulation from marketplace plurai-plugins in plurai-ai/plurai-plugins. Commands read from the repo's manifest — yesterday.

Skills (35)

• analyze-simulation

  Quality gate for a built simulation. Independent critic sub-agents rank root causes across scenarios, harness, evals and RAG; you fix suite-side ones and loop until a critic comes back clean. Driven by /simulation-generate-suite, or run directly. Not the agent-repair step (/simulation-fixes).

• analyze-traces

  Ground the simulation in real production traces (one stage of `/simulate`; callable alone). The SDK exports them locally; the backend turns them into a brief. By default it then turns the brief's trace failures into trace scenarios. Optional; run after /simulation-scan-agent.

• author-trace-adapter

  Author a declarative YAML adapter for a trace format the SDK converter does not know (session logs, custom exporter JSON, raw log lines). Saved under plurai_sim/plurai_adapters/. Run before /trace-to-scenarios for traces that are not LangSmith or LangGraph thread exports.

• build-rag-corpus

  Build a knowledge corpus for an agent's RAG or SQL tools. The SDK extracts chunks locally and uploads them; the backend builds the knowledge graph and test set. One corpus per tool. Run after /simulation-scan-agent for agents with retrieval or text2SQL tools.

• edit-env

  Edit a generated mock environment or a built RAG corpus in place through a multi-turn backend session: change a tool's return shape, seed different world state, fix a stale fact. Run after /generate-mock-env or /build-rag-corpus.

• edit-evals

  Iterate on the generated policies_config.yaml in a backend session: add, change or remove policies, thresholds and judge prompts. Run after /generate-evals.

• edit-scenarios

  Fix a scenario's grading ground truth in place (a wrong expected_answer, stale evidence quotes, a mis-stated target), one field per call. Not for persona, questions or world: regenerate for those.

• extract-eval-intent

  Write eval_intent.yaml, what the customer cares about, from the PRD and scenario variables after 3-4 questions. Optional. Run standalone before a suite build, never inside /simulation-generate-suite.

• extract-scenario-variables

  Write scenario_variables.yaml, the per-agent context every scenario generation reads (purpose, tools, entities, user archetypes, complications). Run after /simulation-scan-agent and before /generate-scenarios. Internal building block.

• generate-evals

  Generate policies_config.yaml by calling the backend eval generation service with the agent's artifacts. Run after /simulation-scan-agent and /extract-scenario-variables.

• generate-mock-env

  Generate and deploy the agent's mock environment with a backend job, then store its env_id and mock url. Normally invoked by /simulation-generate-suite; run alone to rebuild the world after /simulation-scan-agent.

• generate-scenarios

  Generate scenarios on the Plurai-hosted backend (no local LLM credentials required) and persist them as ground truth. The plugin writes only a one-line `plurai_sim/.scenario_set_id` reference; the actual scenarios live on the backend until you delete them.

• ground-scenarios

  Route real material (production conversations, real records, or both) to /trace-to-scenarios or /upload-scenario-bundle, then hand off to /generate-scenarios. Run after /simulation-scan-agent when scenarios should come from real data.

• harness-setup

  Log in to Plurai through the browser, install the SDK, connect the Plurai tools and print what this harness needs. The Plurai commands run it themselves when a login or connection is missing. Run it directly to log in again or to see which account is saved.

• list-scenarios

  List the scenario sets stored on the backend, or inspect one in detail. Read-only — does not run anything. Useful for sanity-checking what scenarios are available before launching a simulation.

• manage-envs

  Save, switch between and list named simulation environments under plurai_sim/envs/. Triggered by natural language such as "save this as env <name>", "switch to env <name>" and "list envs"; there are no slash commands.

• optimize-eval

  Improve one generated llm_as_judge policy: the backend rewrites its rubric or decomposes it into a hybrid judge and benchmarks the candidates; adopt one only on your explicit confirmation. Supports --revert. Run after /generate-evals.

• plurai-diagnose

  Print a bundle to send to Plurai support.

• plurai-uninstall

  Remove what the Plurai plugin created in a project and in your home folder, after you confirm each part. Run it before removing the plugin itself.

• record-regression

  Record a confirmed agent issue as a regression folder under plurai_sim/regressions/ that replays the conversation which exposed it. One backend scoring call measures how often it recurs, kept in replays.json. Run by /simulation-fixes and /simulation-generate-suite, or by hand.

• refine-scenarios

  Interactively steer which scenario types the backend generates, review what changed, then materialize. Alternative to /generate-scenarios; run after /simulation-scan-agent and /extract-scenario-variables.

• simulate

  Simulate your agent end to end (scan, traces, suite, run, fix plan) with one live overview page and feedback in chat. Use it for first and repeat runs; it asks one start question. User-only entry point; each stage is also callable alone.

• simulation-diagnose

  Diagnose one already-failed run directory with error.json files, isolating a local cause (simulate.py, scenarios, config) from a remote one (mock service, seed data). Used by /simulation-fixes; not for a general run-and-diagnose request.

• simulation-fixes

  Diagnose a run's failures and write a ranked fix plan and fixes.md with evidence and code locations; never edits the agent. Records an accepted agent fix with /record-regression BEFORE applying the fix. Stage 5 of /simulate; run after /simulation-run. Optional --verify.

• simulation-generate-suite

  Build the suite: mocked world, scenarios and evals, then iterate on quality until the simulation gives sharp signal, and render a report. Stage 3 of /simulate; run after /simulation-scan-agent.

• simulation-run

  Score the agent: a simulated user plays every scenario against it on the mocked world and all evals are scored, optionally recorded as an experiment in LangSmith, Langfuse, Arize or Braintrust. Stage 4 of /simulate. It scores but does not diagnose (use /simulation-fixes).

• simulation-scan-agent

  Map the agent and write its harness: decide which surfaces are mocked or real, show a topology diagram for approval, then generate simulate.py and plurai.config.yaml. Stage 1 of /simulate; run it first, with the path to the agent module.

• simulation-setup-agentforce

  Internal phase of simulation setup for a hosted Salesforce Agentforce agent (LANG is agentforce). Reached only from simulation-setup Step 10. Never invoke directly; use simulation-setup.

• simulation-setup-bridge

  Internal phase of simulation setup for an agent not written in Python, run through a bridge (LANG is node). Reached only from simulation-setup Step 10. Never invoke directly; use /simulation-scan-agent.

• simulation-setup-python

  Internal phase of simulation setup for a Python agent (LANG is python): simulate.py template, mock contract tests, smoke run. Reached only from simulation-setup Step 10. Never invoke directly; use /simulation-scan-agent.

• simulation-setup-skill

  Internal phase of simulation setup for an agent that is a Claude Code plugin command or skill (LANG is claude-skill). Reached only from simulation-setup Step 10. Never invoke directly; use simulation-setup.

• simulation-setup

  Analyze an agent codebase and generate its simulation wrapper (simulate.py, plurai.config.yaml, simulation_prd.md). Use for "set up simulation", "set up plurai", "generate simulate.py". Internal building block; prefer /simulation-scan-agent.

• trace-to-scenarios

  Turn conversation exports into scenarios that resume where the trace ended. You never read or parse a raw trace file. `--from-brief` (chained by /analyze-traces) turns a trace brief's failures into scenarios and reads the converter's own rendering of a trace, to choose the cut.

• upload-doc-examples

  Upload real example documents (invoices, forms, tickets, PDFs) so generated documents follow real formatting. Only chosen examples leave the machine. Run after /simulation-scan-agent when scenarios involve documents.

• upload-scenario-bundle

  Ground scenario preconditions in real records: pick records locally, one folder per scenario with a ROLES.md, and upload only those. /generate-scenarios then builds from the folders. Run after /simulation-scan-agent.

MCP servers (1)

• plurai

  connects to: ${PLURAI_URL:-https://sandbox.plurai.ai/v1}/mcp

Hooks (22)

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude --bundle regen_gate,command_run_id,halt_gate

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" skill -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude flow_autoarm

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude ask_guard

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" token -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude plugin_root

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" token -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude plugin_root

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude cron_gate

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude cron_gate

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude shell_syntax_guard

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude placeholder_guard

• PostToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude regen_spent

• PostToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude pending_register

• PostToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude ask_answered

• PostToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude failure_notice

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude --bundle session_glossary,session_pin

• UserPromptSubmit

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" prompt -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude --bundle flow_arm,pending_register,flow_heartbeat

• SessionEnd

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude --bundle flow_cleanup,session_pin

• SubagentStart

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude pending_register

• SubagentStop

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude pending_register

• Stop

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude report_guard

• StopFailure

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude flow_interrupt

• PermissionDenied

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude denial_watch

• PostToolUseFailure

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh" -- python3 "${CLAUDE_PLUGIN_ROOT}/hooks/run.py" --harness claude failure_notice

Similar plugins