apsy

Plugin apsy from marketplace auto-psynet in Haoyu-Hu/auto-psynet. Commands read from the repo's manifest — 1 month ago.

Skills (32)

• add-recipe

  Add a new PsyNet recipe file under skills/psynet/psynet-function/ (a new paradigm or cross-cutting capability) and auto-update the parent index in skills/psynet/SKILL.md. Use when extending the PsyNet knowledge pack with something not yet covered — e.g. a new TrialMaker subclass, a new prescreen, a

• analysis-plan

  FORMULATE step — lock the preregistered analysis plan (statistical model, primary/secondary outcomes, decision rules) BEFORE data. This is the holdout that defends against p-hacking.

• auto

  Smart router — single natural-language entry point that picks the right /apsy:* command. Use when the user types /apsy:auto, says 'just do X', or has a request that doesn't obviously map to one specific /apsy:* command. Routing is deterministic + transparent — no LLM-scored guesswork.

• data-quality

  ANALYZE gate G5 — screen the data for quality: completion, attention/manipulation checks, the preregistered exclusions, duplicates/bots, and target-N. Produces the clean analysis dataset.

• design

  FORMULATE step — choose the experimental paradigm and design. Maps the question archetype to a PsyNet paradigm via config/affinity.yaml, reads the recipe + domain priors, may propose elevating to a differentiating paradigm, and specifies conditions, counterbalancing, prescreens, population/languages

• doctor

  Diagnose the Auto-PsyNet runtime — use when the user says 'apsy doctor', 'check my setup', 'is everything working', or before first deploy. Validates that essential dependencies (psynet, dallinger) are installed + detectable, plus Docker/Postgres/Redis, LLM keys, AWS, config.

• export-data

  ANALYZE step — export the experiment's data (psynet export) or load a prior/pilot export, and locate the per-class CSVs for analysis. Also consumes LLM-pilot synthetic data.

• formulate

  FORMULATE step 1 — turn a raw research idea into a structured question, falsifiable hypotheses, and operationalized variables (IV/DV → measures/stimuli). Use at the start of a new experiment or when the user runs /apsy:idea.

• generate-stimuli

  BUILD step (as needed) — create, source, or organize the stimulus set, or wire synth_function for generative paradigms (e.g. media-GSP); register media as PsyNet Assets.

• implement-paradigm

  BUILD step — fill the scaffolded experiment.py with the Trial/Node/TrialMaker code for the chosen paradigm, from its psynet-function recipe and the verified plan. Enforces the PsyNet gotchas and wires bot_response.

• install

  Install Auto-PsyNet's essential dependencies — PsyNet + Dallinger (with optional version pinning) and the Python stats stack — into a chosen Python interpreter. Use when /apsy:doctor reports psynet missing, when the user says 'install apsy/psynet', or after first cloning the plugin. Auto-detects ven

• interpret

  ANALYZE gate G6 — map the executed results to the hypotheses: effect sizes, uncertainty, robustness, and whether H1 is supported per the preregistered decision rules. Verifies the analysis matched the plan.

• iterate

  ANALYZE gate G7 — decide ship or iterate. If iterate, specify the single change + rationale and loop back to BUILD/PILOT, recording the trail. This is the 'improve until satisfied' decision.

• literature-ground

  FORMULATE step — situate the idea in the literature: prior paradigms, expected effect sizes, and the novelty gap. Uses web search + arXiv, plus the Semantic Scholar API when SEMANTIC_SCHOLAR_API_KEY is set.

• llm-pilot

  PILOT gate G3 — run the experiment with LLM-agent participants to validate the pipeline end-to-end, produce synthetic data, exercise the analysis pipeline, and sanity-check that the task is doable. No human spend.

• localize

  Translate the experiment for cross-cultural / multilingual runs. Use when the research plan declares >1 locale, when running a global Lucid panel, or when measurement invariance is part of the G6 rubric. Wraps user-facing strings with the PsyNet translator (`_` / `_p`), runs `psynet translate` to ex

• plan-review

  FORMULATE gate G1 — score the research plan against the methodological + ethics rubric, surface confounds/validity/ethics issues, and decide pass or revise. Run after the plan is drafted.

• power-analysis

  FORMULATE step — compute the required sample size for the planned effect and model. Runs real power computation (bin/apsy-power.py); never guesses N.

• project-dir

  Set or inspect the Auto-PsyNet project directory — the consistent root where new experiments are scaffolded. Use when the user says 'set project dir', 'change project directory', 'where do experiments live', or to relocate an existing project home. Records APSY_PROJECT_DIR in ~/.auto-psynet/config.

• psynet

  PsyNet function knowledge — paradigm recipes (how to design/build each PsyNet experiment type) + cross-cutting capabilities (synchronization, i18n, prescreening, consent, demography, controls, assets, CLI). Use during design/build of any PsyNet experiment; open the matching psynet-function/ subfile.

• recruit

  DEPLOY step — configure, launch, and monitor participant recruitment (Prolific/Lucid/MTurk) within the approved spend cap; run live data-quality and decide extend/pause/stop.

• repro-package

  PUBLISH step — bundle an OSF-ready reproducibility package: code + locked preregistration + analysis scripts/results + anonymized data + manifest. Optional OSF push.

• run-analysis

  ANALYZE step — execute the preregistered analysis EXACTLY (the locked §6 model) on the clean data with real stats; compute effect sizes + CIs and figures. Never narrate numbers.

• run-deploy

  PILOT&DEPLOY gate G4 — deploy the experiment for REAL human data collection. HARD gate: human approval + Cornell IRB attestation + spend cap + G2/G3 green. Never auto-passed at any autonomy level.

• run

  Autonomous research pipeline — spec in → paper out. Walks FORMULATE → BUILD → PILOT → [DEPLOY] → ANALYZE → PUBLISH honoring autonomy_level; HARD G4 (real deploy) always pauses. Synthetic-only by default; --with-deployment adds the real-human Track B branch.

• scaffold

  BUILD step 1 — scaffold the PsyNet experiment project from the verified plan: experiment.py (from the paradigm template), config.txt, requirements.txt, and PsyNet boilerplate. Requires a G1-verified plan.

• setup

  Configure Auto-PsyNet — use on first run or when the user says 'apsy setup', 'configure apsy', or has no ~/.auto-psynet/config. Sets the LLM-participant backend, username/server prefix, AWS/EC2 + base domain, and notes the consent default.

• status

  Report where the current Auto-PsyNet experiment stands — use when the user says 'apsy status', 'where am I', 'what's next', or resumes work. Reads <experiment>/.apsy/state.json.

• test-experiment

  BUILD gate G2 — make the experiment self-verifying with bots and run `psynet test local` until green. Sets test_n_bots + test_check_bot assertions.

• update

  Upgrade PsyNet and/or Dallinger in the active Python environment to a specified or the latest version. Use when the user says 'update psynet', 'upgrade dallinger', 'bump deps to latest', or after /apsy:doctor reports an outdated version. Reports old → new version diff; never auto-upgrades inside an

• wire-timeline

  BUILD step — assemble the full participant timeline: consent → instructions → prescreens → demography → trial maker(s) → feedback → debrief, composing PsyNet's built-in modules. Consent defaults to MainConsent (override via apsy:consent).

• write-paper

  PUBLISH step — assemble a publication-ready paper draft: Methods from the actual pipeline + preregistration, Results from the executed analysis (results.json), Intro/Discussion from the literature + findings, with figures and a required AI-involvement disclosure.

Slash commands (16)

• /apsy:analyze

  Export data and run the preregistered analysis; report effects, then iterate or ship (gates G5/G6/G7).

• /apsy:build

  Generate the PsyNet experiment from a G1-verified plan — scaffold, implement, wire timeline, bot-test to G2.

• /apsy:consent

  Configure the consent form (default PsyNet MainConsent) — separate file? class/function to import? how to use it?

• /apsy:debug

  Run the current experiment for debugging — launches `bin/apsy-debug.sh` in the background, monitors the boot log, and reports when the server is up. Targets: local | ec2 | stop.

• /apsy:deploy

  Deploy for REAL human data collection (gate G4) + recruit. HARD gate — human approval + IRB attestation + spend cap.

• /apsy:export

  Export experiment data from a running `psynet debug local`. Thin wrapper around `bin/apsy-export.sh` that auto-redirects to `$APSY_PROJECT_DIR/data/<study>/` via `--path` (or psynet default `~/psynet-data/export/` if no project dir).

• /apsy:help

  List every `/apsy:*` command with a one-line description, or print detailed help for one command (`/apsy:help <name>`). Also supports `/apsy:help --search <query>` to filter by keyword.

• /apsy:idea

  Start FORMULATE — turn a research idea into a verified, preregistered research plan (gate G1).

• /apsy:lucid

  Configure Lucid/Cint recruitment params (global panels) for the current experiment.

• /apsy:mturk

  Configure Amazon Mechanical Turk recruitment params for the current experiment.

• /apsy:paper

  PUBLISH — assemble the paper draft (Methods from pipeline, Results from analysis) + an OSF-ready reproducibility package.

• /apsy:pilot

  Run the experiment with LLM-agent participants — validate the pipeline + analysis on synthetic data (gate G3). No human spend.

• /apsy:prolific

  Configure Prolific recruitment params (base_payment, wage_per_hour, estimated minutes, qualifications) for the current experiment.

• /apsy:region

  Override the default AWS region for EC2 provisioning (default us-east-1).

• /apsy:services

  Start / stop / status of the runtime services (Redis + PostgreSQL) that `psynet debug local` requires. Thin wrapper around `bin/apsy-services.sh` that surfaces the engine's idempotent operations + auto-setup of the dallinger user + database. Subcommands: status (default) · start · stop · restart.

• /apsy:type

  Override the EC2 instance type for provisioning (default auto-sized m7i.{N}xlarge).

Subagents (9)

• adversarial-reviewer

  Adversarial reviewer (red team) for research plans and findings — surfaces confounds, validity threats, statistical issues, alternative explanations, generalizability concerns, and methodological pitfalls. Optional independent pass at G1 (plan) and G6 (findings). Adversarial but fair; never invents

• code-reviewer

  Code reviewer for generated PsyNet experiments and analysis scripts — enforces the 8 PsyNet gotchas, Python correctness, secret/PII hygiene, and analysis-script-matches-preregistration. Severity-tiered findings (🛑 critical / ⚠️ major / 🔵 minor). Optional pass after implement-paradigm/wire-timeline

• data-analyst

  Data analyst for online behavioral experiments — live data-quality monitoring during recruitment and analysis support (completion, exclusions, attention checks, bot/bad-actor detection, wrangling, visualization). Use during recruitment (Track B) and ANALYZE.

• debugger

  Failure-driven debugger for Auto-PsyNet runs — diagnoses errors, test failures, and unexpected behavior, applies the MINIMAL fix, and verifies. Invoked when `psynet test local` (G2) fails, the LLM pilot crashes, the analysis script errors, or any engine step exits non-zero. Never claims a fix withou

• literature-scholar

  Literature scholar for behavioral-science research — finds and synthesizes prior paradigms, expected effect sizes, the novelty gap, and methodological pitfalls in prior work. Cites only real sources. Use during FORMULATE (apsy:literature-ground) and PUBLISH (Introduction / Discussion sourcing).

• methodologist

  Experimental-design methodologist for online behavioral experiments — paradigm selection, construct/internal/external validity, confounds, counterbalancing, prescreening, and research ethics. Use during FORMULATE and at gate G1.

• psynet-engineer

  PsyNet/Dallinger code-generation engineer — paradigm implementation (Trial/Node/TrialMaker), timeline, controls, assets, bots, the LLM-participant driver, i18n, and the psynet CLI. Owns the paradigm recipe library. Use during BUILD and PILOT.

• science-writer

  Academic science writer for behavioral-research papers — Methods (from the actual pipeline), Results (from executed analyses only), Intro/Discussion (from literature + findings), figures, reproducibility, and required AI-involvement disclosure. Use during PUBLISH.

• statistician

  Statistician/psychometrician for behavioral experiments — power analysis, preregistered analysis plans, model choice, multiple comparisons, effect sizes, measurement invariance, and mixed human-AI-sample modeling. Use during FORMULATE (power + analysis plan) and ANALYZE.

Hooks (4)

• SessionStart

  ${CLAUDE_PLUGIN_ROOT}/hooks/load-experiment-context.sh

• SessionStart

  ${CLAUDE_PLUGIN_ROOT}/hooks/first-run-nudge.sh

• PreToolUse

  ${CLAUDE_PLUGIN_ROOT}/hooks/psynet-lint.sh

• PreToolUse

  ${CLAUDE_PLUGIN_ROOT}/hooks/spend-gate.sh

Similar plugins