How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

scripts - Skill - OpenSmartRoute

Name: scripts
Author: MohammadErfan-Jabbari

Imported from MohammadErfan-Jabbari/brain-alignment (scripts/AGENTS.md). Install upstream with npx skills add MohammadErfan-Jabbari/brain-alignment --skill scripts. Copyright stays with the author.

Scripts Agent Guidance

This folder holds experiment runners, data adapters, analysis scripts, literature-sweep utilities, and figure builders. Treat scripts as part of the research evidence pipeline, not standalone product code.

Rules

Do not create new README.md files under scripts/; folder-local agent instructions belong in AGENTS.md.
Run Python with uv run.
Use HF_HOME=/home/centcom/data/hf-cache when loading cached Hugging Face models.
Before changing or running an experiment script, read the matching docs/experiments/E*.md and current status in docs/status.md.
Put heavy outputs in gitignored outputs/ or data/, not in tracked source directories.
E028 is a narrow exception: its synthetic-development runner and analyzer intentionally reject repository outputs/ and data/. Write their ephemeral, non-load-bearing artifacts to a fresh safe temporary directory outside the repository.
Preserve anti-artifact controls: contiguous splits, nuisance baselines, matched budget/perplexity where the design requires them, capacity-fair feature comparisons, and explicit seed handling.
A synthetic run validates plumbing only. Do not present synthetic numbers as science.

Common Paths

Path	Use
`pilot_lib.py`	Shared experiment primitives: seeding, hidden states, baselines, contiguous CV, ridge encoding, variance partition.
`data_adapters.py` / `lebel_adapter.py`	Dataset loading and normalization.
`run_*.py`	Experiment runners.
`analyze_.py` / `reanalyze_.py`	Post-run analysis and robustness checks.
`e028_vaidya_crossmodal_falsification.py`	Outcome-blind E028 synthetic Stage-1 runner. Its only commands are `manifest`, `synthetic-replay`, `selftest`, `benchmark`, and `bundle`; every output remains explicitly not endpoint-ready.
`e028_analyze_vaidya_stage1.py`	Fail-closed analyzer for the E028 synthetic Stage-1 schema and exact patient-level sign-flip gate. It never licenses Stage 2 or accepts a neural endpoint.
`build_text_feature_target_cache.py`	E016 matched-information control target builder: frozen LM text features projected into the Phase-3 target-cache schema; pair with `run_tribe_phase3.py --target-label textfeat`.
`run_tribe_phase3.py`	E016 matched-budget runner for KD-only / target-MSE / target-permuted arms. Use optional `--save-model-dir` on future post-positive/control runs when trained students may be needed for real-brain or probe follow-up; saved artifacts are gitignored and are not verdicts.
`e016_eval_saved_student_alignment.py`	Post-hoc E016 helper that loads saved `model_artifact_dir` students and scores them on the real Tuckute alignment endpoint with the E003 contiguous-CV nuisance-subtracted protocol. Use only after artifacts exist; outputs are diagnostics for `/interpret`, not verdicts.
`e016_analyze_tuckute_alignment.py`	Post-hoc E016 helper that reads one or more saved-student Tuckute alignment JSONs, aligns complete TRIBE/textfeat seed grids, and writes seed-level real-brain contrasts/PCA robustness. It is a diagnostic handoff for `/interpret`, not a verdict engine.
`e016_audit_tuckute_alignment.py`	Local audit helper that recomputes a saved-student Tuckute analysis from raw alignment rows and checks row/protocol/arithmetic/PCA consistency before `/interpret`. It is not an independent evaluator and not a verdict engine.
`e016_make_textfeat_control_script.py`	Writes the post-E016 full text-feature control launcher under `outputs/`; use after a positive TRIBE Phase-3 result makes the matched-information control necessary.
`analyze_tribe_phase3.py`	E016 Phase-3 gate/analyzer; writes conservative readiness fields, target-cache metadata, and `paper_branch_hint`. The hint routes to the paper plan only after `science_ready=true`; it is not a verdict or rung flip.
`e016_compare_target_controls.py`	Post-positive E016 comparator for ready TRIBE and text-feature analyzer JSONs; compares within-target paired gains over KD/permuted controls and emits target-scope metadata, a branch hint, and top-venue reviewer-burden metadata, never a verdict.
`manuscript_check.py`	Deterministic manuscript gate: evidence-marker resolution, keyed-number resolution across the `\input` graph, bare result-like numbers, `\gap` survival under `--share-ready`, and the LaTeX build. Builds with latexmk only, and judges undefined references from the final `.log` rather than latexmk's multi-pass stdout; both matter, see `docs/manuscript/AGENTS.md`. Moved here from `.claude/` by D068.
`prose_lint.py`	Deterministic prose floor for banned tokens. A token list is complete by construction; register and semantics are not lintable and belong to review (L052).
`figures/`	Figure generation scripts. `figures/make_brain_figure.py` builds the defense-deck LH language-network figure (EvLab SN220 parcels on fsaverage pial) into `outputs/presentations/brain-lh-language/`, downloading the parcel volume to `data/evlab-parcels/` on first run. `figures/make_tribe_response_figure.py` renders one TRIBE v2 synthetic response (E016 phase-3 route, `predict` stage in the TRIBE venv, `render` stage in the thesis env) on fsaverage5 into `outputs/presentations/tribe-response/`.
`litsweep/`	Literature search and PDF utilities. See `litsweep/AGENTS.md`.

Toy-pilot code (E001)

The scientific question, the estimand rule, the anti-confound protocol, and the run commands belong to E001, which owns them. This section is navigation only.

File	Role
`pilot_lib.py`	Scientific spine: seeding, hidden-state extraction, scalar and static nuisance baselines, contiguous-block CV, ridge encoding, capacity-fair variance partition.
`distill.py`	GPT-2-medium to GPT-2 distillation loop: logit KD plus an optional trainable brain-alignment head.
`data_adapters.py`	Common `(texts, fmri, meta)` interface with synthetic, Pereira sentence, and Pereira neural-response paths.
`run_toy_pilot.py`	Entry point; writes `outputs/toy-pilot/<run>/{summary.json,results.csv}`.
`../configs/toy_pilot.json`	Hyperparameters and smoke overrides.

Install scripts and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "mohammaderfan-jabbari-brain-alignment-scripts", "kind": "skill", "name": "scripts", "description": "This folder holds experiment runners, data adapters, analysis scripts, literature-sweep utilities, and figure builders. Treat scripts as part of the research evidence pipeline, not standalone product code.", "publisher": "MohammadErfan-Jabbari", "version": "1.0.0", "capabilities": { "domains": [ "coding" ], "tags": [ "skill-md", "reference", "onboarding", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "This folder holds experiment runners, data adapters, analysis scripts, literature-sweep utilities, and figure builders. Treat scripts as part of the research evidence pipeline, not standalone product code." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/MohammadErfan-Jabbari/brain-alignment", "path": "scripts/AGENTS.md", "ref": "3d7341876547d6e54ab826a79edd5eb25b664ae2", "url": "https://github.com/MohammadErfan-Jabbari/brain-alignment/blob/3d7341876547d6e54ab826a79edd5eb25b664ae2/scripts/AGENTS.md", "key": "MohammadErfan-Jabbari/brain-alignment/scripts/AGENTS.md" } }, "instructions": "# Scripts Agent Guidance\n\nThis folder holds experiment runners, data adapters, analysis scripts, literature-sweep utilities, and figure builders. Treat scripts as part of the research evidence pipeline, not standalone product code.\n\n## Rules\n\n- Do not create new `README.md` files under `scripts/`; folder-local agent instructions belong in `AGENTS.md`.\n- Run Python with `uv run`.\n- Use `HF_HOME=/home/centcom/data/hf-cache` when loading cached Hugging Face models.\n- Before changing or running an experiment script, read the matching `docs/experiments/E*.md` and current status in `docs/status.md`.", "cost": { "context_tokens": 1556 } }

scripts

About

Scripts Agent Guidance

Rules

Common Paths

Toy-pilot code (E001)

Install scripts and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews