text stringlengths 3 8.33k | repo stringclasses 52
values | path stringlengths 6 141 | language stringclasses 35
values | sha stringlengths 64 64 | chunk_index int32 0 273 | n_tokens int32 1 896 |
|---|---|---|---|---|---|---|
/////////////////////////
//Discord Bot für 1AHIF//
/////////////////////////
const Discord = require('discord.js');
const bot = new Discord.Client();
const prefix = '.';
bot.on('ready', () => {
console.log('AHIF-BOT ist jetzt online!');
bot.user.setActivity('1AHIF', { type: "WATCHING"}).catch(console.error... | 1AHIF-BOT | main.js | JavaScript | 71c4053871fb1914ac32ddfdbdab85a034140f6f44209f908b6b3f4b4e5ba588 | 0 | 452 |
{
"name": "1ahif-bot",
"version": "1.0.0",
"description": "",
"main": "main.js",
"scripts": {
"test": "echo \"Error: no test specified\" && exit 1",
"start": "node main.js"
},
"author": "Julian Schmidt",
"license": "ISC",
"dependencies": {
"discord.js": "^12.3.1",
"webuntis": "^1.12.0"... | 1AHIF-BOT | package.json | JSON | 927bfac04090d59e1be821747d8f2d4af5e19e1b9d60e09eeb7bc0c27d2cef49 | 0 | 124 |
Worker: node main.js | 1AHIF-BOT | procfile | Procfile | bc442796387a3b85efab858c1e22784cff1c22826ce8de6148a4f384cbfb06f8 | 0 | 6 |
# Auto detect text files and perform LF normalization
* text=auto
| adapter-lab | .gitattributes | Git Attributes | 1a1dbe176bc233b499d35a57db7513f2941c99ab9759f177830c9149be99005b | 0 | 13 |
__pycache__/
*.py[cod]
.pytest_cache/
.ruff_cache/
.venv/
| adapter-lab | .gitignore | Git Ignore | 160d84bae49f571dbe2778651a39f0e633b464bd0a6587c9156127ba9b4c2368 | 0 | 17 |
<!-- GSD:project-start source:PROJECT.md -->
## Project
**adapter-lab**
`adapter-lab` is a single-tenant, self-contained, offline-capable enterprise lab for producing signed, auditable LoRA and QLoRA adapters from open-weight base models. It lets regulated organizations fine-tune LLMs on sensitive tenant-controlled ... | adapter-lab | AGENTS.md | Markdown | 3e7d6096df2de1c904221b5345e21f73f4541faec6cfdb6073b3852b11fec334 | 0 | 896 |
standardize on it.
- **PII:** Microsoft Presidio Analyzer/Anonymizer with YAML-loaded tenant recognizers and plugin entry points.
- **Evaluation:** `lm-evaluation-harness` as the default capability harness; separate plugins for BFCL, tau-bench, safety panels, and tenant custom suites.
- **Provenance/signing:** in-toto ... | adapter-lab | AGENTS.md | Markdown | 9af44c1c12ddfe7bfc9baf140b2518e478bbb53ad2453ac4a63eaccb3f99489d | 1 | 896 |
- Do not make DeepSpeed the only distributed backend; PyTorch FSDP is the more general default.
- Do not put tenant plugins in the corpus directory or load arbitrary code from input data paths.
- Do not store PII values, canary text, signing secrets, or HSM PINs in logs, manifests, or model cards.
- Do not produce prop... | adapter-lab | AGENTS.md | Markdown | e9cdf4d25c62323c4de2591b60c54f8b2505eb612ec66bc726880b31c8dda660 | 2 | 896 |
.md` first if it exists; otherwise read `.planning/graphs/GRAPH_REPORT.md`.
- If `graphify-out/wiki/index.md` exists, navigate it before reading raw files.
- For cross-module "how does X relate to Y" questions, prefer `node "$HOME/.codex/get-shit-done/bin/gsd-tools.cjs" graphify query "<term>"`, `graphify query "<quest... | adapter-lab | AGENTS.md | Markdown | fd26906abb79f60b5ebb93705701b5deeda451e17240c43736460efa3ae744cf | 3 | 244 |
[project]
name = "adapter-lab"
version = "0.1.0"
description = "Offline-capable enterprise lab for signed, auditable LoRA and QLoRA adapter production."
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
"pyarrow>=16,<23",
"pydantic>=2.12,<3",
"pyyaml>=6.0,<7",
"typer>=0.20,<1",
]
[projec... | adapter-lab | pyproject.toml | TOML | 670ca33fa8f2a200880a1820256736cf8bc359fb46f34cd14204a20a7ec56008 | 0 | 251 |
# adapter-lab
Offline-capable enterprise lab for producing signed, auditable LoRA and QLoRA adapters from local open-weight base models.
## Foundation CLI contract
`adapterlab` is a thin Typer CLI over importable SDK services. Durable behavior belongs in
`adapterlab.sdk`; CLI commands translate operator input, call ... | adapter-lab | README.md | Markdown | c8190ad93d81ebb7ed64dacc6589f22287eabb6061607d5989b36c9e8f732790 | 0 | 889 |
# Product Requirements Document — `adapter-lab`
**Version:** 0.9 (Draft for Implementation)
**Status:** Engineering reference, pre-implementation
**Audience:** AI agents, ML/MLOps engineers, security engineers, applied researchers
**Reference deployment:** Bundesrechenzentrum (BRZ) sovereign LLMaaS (illustrative, not ... | adapter-lab | vision.md | Markdown | 6e9bb80fc0fdb2b5a90886d1be9363a74b72584b032ea36a4b0ee6942361f752 | 0 | 896 |
─────────────────────────────────────┐
│ adapter-lab Control Plane │
│ (CLI, Python SDK, run-orchestrator) │
└─────────────────┬────────────────────────┘
│ manifests, run-state
┌─────────... | adapter-lab | vision.md | Markdown | 23e7dc27d48e47979922f10306a0298e94eae9ccda7bb0308d10622e57e76d17 | 1 | 896 |
the lab itself (code, base models, public benchmark data) is reproduced from a pinned, signed software bill of materials (SBOM).
- **Artifact boundary:** outputs are signed; downstream consumers (serving stack, regulator) verify signatures before loading.
---
## 3. Subsystem Specifications
### 3.1 Ingest
**Purpose:... | adapter-lab | vision.md | Markdown | c365e41b7ddc90fd129522ea018dc92e5bb94f23d9ff9b70f3de792b902a4331 | 2 | 896 |
`synth.yaml` selecting strategies, generators (which open-weight model to use as the "teacher"), sampling params, quality filters.
- Optional `seeds.jsonl` of human-written examples (for Self-Instruct bootstrap).
**Supported strategies:**
| Strategy | Origin | Mechanism | When to use |
|---|---|---|---|
| **Magpie** ... | adapter-lab | vision.md | Markdown | 08cff5062fdab5dfe2d99bed202ab31972d7ed655c06edea8ca323cb91d7b28b | 3 | 896 |
. model SHA and license, per-strategy counts, filter pass rates, seed/RNG state).
### 3.3 Train
**Purpose:** Produce a PEFT-compatible adapter from a frozen base model using the chosen training method.
**Inputs:**
- `sft.jsonl` and/or `pref.jsonl`.
- `train.yaml` (training method, hyperparameters, optimiser, FSDP/De... | adapter-lab | vision.md | Markdown | 6de98494a0f4142d146033a909d8cf1bc45fccef33c7d55e7c819493652d419f | 4 | 896 |
language adaptations available in the harness. | EleutherAI |
| **IFEval** | Instruction following via 541 prompts (25 instruction types) such as "write in more than 400 words" or "mention the keyword of AI at least 3 times". | Zhou et al. 2023, arXiv 2311.07911 |
| **BFCL (v3 / v4)** | Function-calling and agentic eva... | adapter-lab | vision.md | Markdown | 15a10369b4cca751aa9efbd095fd8215468f5e485e43042ec93d33b2474e6496 | 5 | 896 |
-style).
- **Metric:** AUC and **TPR at 1% FPR** (the standard low-FPR membership-inference figure used across the LLM MIA literature, e.g., He et al. USENIX Security 2025).
- **Pass criterion (default):** AUC ≤ 0.55 AND TPR@1%FPR ≤ 0.05. Tenants tighten or loosen via `verify.yaml`.
#### 3.5.2 Data extraction / canary... | adapter-lab | vision.md | Markdown | da8d32530e45e33c259616c5555257e356718e5f9119b33cde73684e2700984f | 6 | 896 |
run the OR-Bench-Hard-1K subset (1,000 prompts drawn from the 80,000-prompt full set) plus the 600 toxic anchor prompts; per Cui et al. (arXiv 2405.20947) "OR-Bench comprises 80,000 over-refusal prompts across 10 common rejection categories, a subset of around 1,000 hard prompts that are challenging even for state-of-t... | adapter-lab | vision.md | Markdown | a1563c753ecdf660542f3cce23d50a760d4ce3e1685d40e7e5c45aa50211c139 | 7 | 896 |
` with a YAML front matter. Per the HF Hub docs, "A model repo will render its README.md as a model card. The model card is a Markdown file, with a YAML section at the top that contains metadata about the model." The card follows the 9 canonical sections from Mitchell, Margaret et al., "Model Cards for Model Reporting,... | adapter-lab | vision.md | Markdown | 3fea459760578a1d0186bf89564ee0ba63d5b62e9720af52854699a59e269cfa | 8 | 896 |
---|
| LoRA | 16 (bf16) | AdamW / Adafactor | `r`, `alpha`, `dropout`, `target_modules`, `modules_to_save` | Baseline |
| QLoRA | 4 (NF4) | paged AdamW 8-bit | + `bnb_4bit_quant_type=nf4`, `bnb_4bit_use_double_quant=true`, `bnb_4bit_compute_dtype=bf16` | Default (Dettmers et al., NeurIPS 2023) |
| 8-bit LoRA | 8 | Adam... | adapter-lab | vision.md | Markdown | 7038004c237755479045c6e1ffeb08060d8f73102d00e8a27699fd2b51c6b159 | 9 | 896 |
}` triple, not by name alone.
- Datasets are content-addressed; any change to a dataset row changes the dataset hash.
- A `repro/` directory in the bundle contains the exact configs used.
- Deterministic mode is opt-in (slower); statistical reproducibility (within ±ε on eval metrics) is the default.
### 9.2 Provenance... | adapter-lab | vision.md | Markdown | 608734a2aee7a4a8d48604496572d9e4e9994c22b325605d4f00109edac4d02b | 10 | 896 |
-color`
- `--profile <name>` (named tenant profiles for multi-environment work)
- `--dry-run` (validate configs, do not execute)
- `--from-manifest <path>` (resume any subsystem from a prior manifest)
Exit codes: `0` success; `2` config error; `3` validation error; `4` gate failure; `5` signing failure; `1` everything... | adapter-lab | vision.md | Markdown | 740aa267d243afaac6ee34298906e7c9a91a5f53b31c05eea9edc38002b335fc | 11 | 896 |
- [ ] Implement reward-model–based preference generation for DPO (FsfairX-LLaMA3-RM-v0.1 default; tenant-pluggable).
- [ ] Implement filter pipeline; insert canaries deterministically given run seed.
- [ ] Tests: assert per-strategy schema conformance, language consistency, dedup behaviour.
### 12.4 Phase 3 — Train
- ... | adapter-lab | vision.md | Markdown | b3cc966bda8bfefa296a67bf4b2a72d0f11bc94383d59fdb79fbb0170e786a13 | 12 | 896 |
7B or Llama-3-8B), DE/EN only, LoRA + QLoRA only, lm-eval-harness + IFEval + one safety panel (XSTest). Acceptance: an operator can reproduce a signed adapter from raw documents with one CLI command.
2. **Stage 2 — Security gates (8 weeks).** Land Phase 5 in `report`/`warn` mode first; flip to `block` only after the ga... | adapter-lab | vision.md | Markdown | 54457c74777e43c5cf6b5a1d068af6b554b13e6a25ef06db14b0ef2d8ef4616d | 13 | 896 |
benchmarks are strongest in English. DE quality is good; PL/IT/HU quality varies. Tenants MUST sample-review multilingual synth outputs.
- **Reference deployment ≠ requirement.** BRZ is named throughout as the reference tenant. Nothing in this PRD requires BRZ-specific infrastructure (no Artifactory, no Fulcio, no Tekt... | adapter-lab | vision.md | Markdown | d176059459a85f12d9ab4ff1ee4f7cbc2102772486fbc5e324c3aa8e6cf4cea9 | 14 | 109 |
{
"hooks": {
"PreToolUse": [
{
"matcher": "Bash",
"hooks": [
{
"type": "command",
"command": "/Users/julianschmidt/miniconda3/bin/graphify hook-check"
}
]
}
]
}
}
| adapter-lab | .codex/hooks.json | JSON | 723a1b460a2345fcb5dfba362c45e5d09e9b92edf5fbef2e4c84baa4272e3034 | 0 | 59 |
{
"model_profile": "quality",
"commit_docs": true,
"parallelization": true,
"search_gitignored": false,
"brave_search": false,
"firecrawl": false,
"exa_search": false,
"phase_naming": "sequential",
"project_code": null,
"claude_md_path": "./CLAUDE.md",
"git": {
"branching_strategy": "none",
... | adapter-lab | .planning/config.json | JSON | 1869436fbd7510b02054eeff02adea4d39d71610a8d7ab9b164c963f4e76760c | 0 | 717 |
# Milestones
## v1.0 MVP (Shipped: 2026-05-29)
**Phases completed:** 8 phases, 37 plans, 102 tasks
**Key accomplishments:**
- Built the SDK-first runtime foundation: Typer CLI, Pydantic contracts, schema exports, content-addressed artifact store, run DAG contracts, plugin trust records, and safe initialization.
- A... | adapter-lab | .planning/MILESTONES.md | Markdown | b3b8ac0d2871e6e3f1c3c9fef754701be9ca0c2b3f5e9072f9a1fba9afa40da0 | 0 | 338 |
# adapter-lab
## What This Is
`adapter-lab` is a single-tenant, self-contained, offline-capable enterprise lab for producing signed, auditable LoRA and QLoRA adapters from open-weight base models. It lets regulated organizations fine-tune LLMs on sensitive tenant-controlled corpora without exporting data, model artif... | adapter-lab | .planning/PROJECT.md | Markdown | a931219546fca7f8f0c04fcd9820d04fe1a92a076f77502b85abb596d4614380 | 0 | 896 |
SBOMs, and build provenance - v1.0.
### Active
Next milestone requirements are intentionally empty. Define them with `$gsd-new-milestone`.
### Out of Scope
- SaaS hosting or multi-tenant cloud operation - the product is a deployable package with one tenant per instance.
- Serving, routing, or prompt orchestration a... | adapter-lab | .planning/PROJECT.md | Markdown | d9ca113160bb4b6311f0d5856dee6e9734686e23f2f5301f50ec3cdd6ef5c1d3 | 1 | 896 |
-Bench, and tenant custom suites where practical. Reports and review bundles must sanitize raw PII, canaries, and secrets.
Verification is a blocking security layer after evaluation and before signing. It must produce structured PASS/FAIL evidence for membership inference resistance, data extraction and canary recall,... | adapter-lab | .planning/PROJECT.md | Markdown | cdf659a23144c253738a44aac67bf46d4d2fd84cfa8efbdf3e9119bb60cbd2dd | 2 | 896 |
manifest-based artifact handoffs the central architecture | Enables resumability, replacement, audit, and independent verification | Accepted |
| Add a final `run.manifest.json` as the run-level evidence index | Stage manifests alone are not enough for end-to-end audit or resume | Accepted |
| Keep serving out of scope... | adapter-lab | .planning/PROJECT.md | Markdown | cd70e87fd9481fad825a00c10906bd889e4523105e445662820328f764151b69 | 3 | 503 |
# Retrospective
## Milestone: v1.0 - MVP
**Shipped:** 2026-05-29
**Phases:** 8
**Plans:** 37
**Requirements:** 100/100 audited complete
### What Was Built
- SDK-first package foundation with strict CLI JSON contracts, schema-exported Pydantic models, immutable artifact refs, run DAG validation, plugin trust evidenc... | adapter-lab | .planning/RETROSPECTIVE.md | Markdown | 40158b38e68f5392dbb54b4a23bb02d3383efb8424ba5d0d8e98df4747c8bf3d | 0 | 553 |
# Roadmap: adapter-lab
## Milestones
- [x] **v1.0 MVP** - Phases 1-8 shipped on 2026-05-29. Archives: `.planning/milestones/v1.0-ROADMAP.md`, `.planning/milestones/v1.0-REQUIREMENTS.md`, `.planning/milestones/v1.0-MILESTONE-AUDIT.md`.
## Phases
<details>
<summary>v1.0 MVP (Phases 1-8) - SHIPPED 2026-05-29</summary>... | adapter-lab | .planning/ROADMAP.md | Markdown | 057893b317b122b52a312fd1788bf8c98b8f48ef75a78ff52a3ad32a2f319974 | 0 | 640 |
---
gsd_state_version: 1.0
milestone: v1.0
milestone_name: milestone
status: Awaiting next milestone
last_updated: "2026-05-29T11:56:02.470Z"
last_activity: 2026-05-29 — Milestone v1.0 completed and archived
progress:
total_phases: 8
completed_phases: 8
total_plans: 37
completed_plans: 37
percent: 100
---
# ... | adapter-lab | .planning/STATE.md | Markdown | f44f385a4de56ec69cc9d332c5b8052e84a686038b418a9f33476a743bf32738 | 0 | 896 |
Phase 05: Training evidence is aggregate/ref-only; CLI JSON, manifests, metrics, experiments, and adapter metadata omit raw training and preference text.
- Phase 06: Evaluation planning is SDK-first, local-resource gated, manifest-centered, privacy-safe, and fixture-backed by default.
- Phase 06: Evaluation execution a... | adapter-lab | .planning/STATE.md | Markdown | a45c340536a9fc0fba13d41fb707d5506ee16cfe959872045f811a3f87c9210f | 1 | 516 |
# Graph Report - adapter-lab (2026-05-29)
## Corpus Check
- 279 files · ~61,488,291 words
- Verdict: corpus is large enough that graph structure adds value.
## Summary
- 1762 nodes · 3207 edges · 163 communities (133 shown, 30 thin omitted)
- Extraction: 61% EXTRACTED · 39% INFERRED · 0% AMBIGUOUS · INFERRED: 1240 e... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | a89e9ceb354b742426540568930395ef1bdd9ed236b6409dc9b798fc581e794c | 0 | 896 |
[[_COMMUNITY_Community 80|Community 80]]
- [[_COMMUNITY_Community 81|Community 81]]
- [[_COMMUNITY_Community 82|Community 82]]
- [[_COMMUNITY_Community 83|Community 83]]
- [[_COMMUNITY_Community 84|Community 84]]
- [[_COMMUNITY_Community 85|Community 85]]
- [[_COMMUNITY_Community 86|Community 86]]
- [[_COMMUNITY_Commun... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 883b353e4d7cd1686fdf20da5f12cd0b3578c25c0c55eb866841ef09b7e89642 | 1 | 896 |
ConfigError` - 30 edges
8. `ContractValidationError` - 30 edges
9. `BaseModelRef` - 29 edges
10. `run_training()` - 27 edges
## Surprising Connections (you probably didn't know these)
- `test_ingest_outputs_do_not_expose_sensitive_values()` --calls--> `run_ingest()` [INFERRED]
tests/privacy/test_enterprise_outputs.... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 22ae289164e9f13e48b819c97199ad4cd530a53707e6010d72fea38800ef6fa2 | 2 | 896 |
, Load a JSON or YAML contract document from disk., Return supported schema validation kinds., Load a JSON or YAML contract document from disk., Load a JSON or YAML contract document from disk. (+22 more)
### Community 9 - "Community 9"
Cohesion: 0.17
Nodes (14): PiiConfig, PII redaction configuration., load_records()... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | ef0ffa621f23089dcbe76539b472bf06fe5d3c0ccf3c895b52dd1fed97b34684 | 3 | 896 |
Community 35 - "Community 35"
Cohesion: 0.14
Nodes (13): Audit And Logs, Bundle Packing, code:bash (uv run pytest tests/training/test_peft_roundtrip.py -m train), code:text (ingest -> synthesize -> train -> evaluate -> verify -> sign ), Doctor, Identity And RBAC, Independent Bundle Verification, Metrics And Experiments... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 90ef0c3a422214915c21a32fcc40642b9e74061785487bcaab9fa02f53636684 | 4 | 896 |
Sign & Pack, 12.8 Phase 7 — End-to-End & Hardening (+1 more)
### Community 59 - "Community 59"
Cohesion: 0.13
Nodes (30): _emit_success(), evaluate_command(), _fail(), _fail_validation(), _json_enabled(), pack_command(), Typer CLI shell for adapter-lab foundation commands., Generate canonical SFT and preference datase... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 395168f87188b698801c107a53ebec648ad36e84fb6393a519aa9d8e2a03982f | 5 | 896 |
test_canary_evidence_has_no_cleartext_value_field(), test_preference_record_requires_chosen_rejected_and_judge(), test_sft_record_contains_training_and_provenance_fields() (+53 more)
### Community 71 - "Community 71"
Cohesion: 0.22
Nodes (9): Dry-run validate a run DAG without executing stages., Dry-run validate a run... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | a09d48bb20e0030321035e19d798359af53eb594bedad2b4ccf802de97a6d12a | 6 | 896 |
_quantization_for(), Deterministic fixture backend for PEFT-compatible adapter outputs., Write a minimal PEFT-compatible adapter layout through the artifact store., run_fixture_training(), get_method_runner(), Training method registry and shared evidence types., Privacy-safe method execution evidence., Return JSON-safe... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | f6cf7c421c5d37d4cebd891425e3e09620e7a7ac9964f917a97205c964e6d58d | 7 | 896 |
, Verification policy evaluation and signing block helpers., Return a stable SHA-256 hash for a gate policy., Evaluate one metric against one threshold., Return privacy-safe reason codes for failing scores., Evaluate all policy thresholds against a metric payload. (+2 more)
### Community 101 - "Community 101"
Cohesion... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 92da359df4fdc33526d4dfc27d29f72628b6ec946513c012905f11dae97e34b6 | 8 | 896 |
run-level evidence index from stage refs., Persist run.manifest.json through the local artifact store., write_run_manifest(), artifact_ref() (+2 more)
### Community 119 - "Community 119"
Cohesion: 0.24
Nodes (9): Local-first evaluation resource resolution., Resolved resource path plus original ref., Resolve one local ... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | cb025265e499daa57073b44838cec46f647cdf1b101e8082fef755119af32b55 | 9 | 896 |
redact_record(), RedactionResult, _write_sidecar()
### Community 133 - "Community 133"
Cohesion: 0.33
Nodes (4): End-to-end run orchestration service., CLI-safe run orchestration result., Return JSON-safe run status and refs., RunResult
### Community 134 - "Community 134"
Cohesion: 0.33
Nodes (4): generate_model_card... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 86ff1e4eca96e6e3545e09c0e06fd8aa509c4081ba78b4d133ae91364ede8111 | 10 | 896 |
161 - "Community 161"
Cohesion: 0.22
Nodes (8): BuildProvenance, Supply-chain evidence contracts., Reference to a generated SBOM artifact., Local build and dependency provenance metadata., Local wheelhouse and mirror readiness status., SBOMRef, WheelhouseReadiness, test_supply_chain_models_are_strict_enough()
## Knowl... | adapter-lab | .planning/graphs/GRAPH_REPORT.md | Markdown | 632de8c994e1a8fde491e3b33f9e3ac03ff539a9d5d11b9be0f9bc1c113964a8 | 11 | 781 |
---
milestone: v1.0
audited: 2026-05-29T11:54:00Z
status: passed
scores:
requirements: 100/100
phases: 8/8
integration: 8/8
flows: 7/7
gaps:
requirements: []
integration: []
flows: []
tech_debt:
- phase: 05-adapter-training
items:
- Real CUDA QLoRA, 8-bit LoRA, and multi-GPU FSDP execution req... | adapter-lab | .planning/milestones/v1.0-MILESTONE-AUDIT.md | Markdown | a67b5f5b38154302c4ad66e88285a9d58fee83c2338e357bb5c38da2f1f7f16c | 0 | 896 |
PEFT adapter outputs, training manifests, and no-network/privacy lanes. |
| Training -> evaluation | passed | Phase 06 verification covers adapter/base comparison reports, local suite resources, evaluator plugins, manifests, and sanitized report refs. |
| Evaluation -> verification | passed | Phase 07 verification cove... | adapter-lab | .planning/milestones/v1.0-MILESTONE-AUDIT.md | Markdown | 2e866604f0222c7e66a6b3eb2d36965f5f46e213bd4d9c5b3f5cca9d96faec4e | 1 | 697 |
# Requirements Archive: v1.0 MVP
**Archived:** 2026-05-29
**Status:** SHIPPED
For current requirements, see `.planning/REQUIREMENTS.md`.
---
# Requirements: adapter-lab
**Defined:** 2026-05-26
**Core Value:** Enterprise operators can produce, verify, sign, and independently audit sensitive-data LLM adapters entire... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | d5f84c52e33b5fb186754ab918b9ef6b54edef549b8689b35dd051f12ce3ed69 | 0 | 896 |
.
- [x] **ING-06**: Ingest loads predefined and tenant custom Presidio-compatible PII recognizers from configuration.
- [x] **ING-07**: PII redaction is destructive by default and records entity type, span, score, operator, and recognizer metadata without storing raw PII values.
- [x] **ING-08**: Optional reversible re... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 632982fe5e25d30ac3fca7cd860ca0c2d916600233c40c66e0f55bf1bdba9aeb | 1 | 896 |
noise multiplier, and RDP accounting recorded in the manifest.
- [x] **TRN-05**: Training refuses to fetch base weights at training time in airgapped mode.
- [x] **TRN-06**: Base models are referenced by repository/name, revision, weight SHA-256, tokenizer SHA-256, and license, not by mutable name alone.
- [x] **TRN-07... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 371874bf8678fb0b16a88bc2386d046108752aff6e28fc532b810f8bd23ab98c | 2 | 896 |
**: Verification reports MIA AUC and TPR at 1% FPR against configured thresholds.
- [x] **VER-04**: Verification tests canary exposure and verbatim recall without logging canary strings in cleartext.
- [x] **VER-05**: Verification runs prompt-injection resistance checks over configured benchmark families.
- [x] **VER-0... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 2c693608866400980658b0d816b1441afcd27d07bf81e53888703c8f255647ba | 3 | 896 |
-02**: The package emits local metrics dumps and optionally exposes Prometheus-compatible or OpenTelemetry-compatible metrics where deployment mode permits.
- [x] **OPS-03**: The package records experiment metadata, hyperparameters, metrics, and artifact refs in a local MLflow-compatible or Aim-compatible backend.
- [x... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 2dc6f149c9e74bd801588c2b15a390366aa44278dae63821c1bdd69b19099a8d | 4 | 896 |
*ECO-04**: Domain-specific tau-bench extensions can be scaffolded from tenant tool inventories.
## Out of Scope
Explicitly excluded. Documented to prevent scope creep.
| Feature | Reason |
|---------|--------|
| SaaS hosting | Conflicts with single-tenant sovereign and regulated deployment model. |
| Multi-tenant co... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 2b16322bafa7ac73ff5f2857a42d36c9294e0177bdcb9dbe75c167b892d92ff2 | 5 | 896 |
-10 | Phase 5 | Complete |
| TRN-11 | Phase 5 | Complete |
| TRN-12 | Phase 5 | Complete |
| TRN-13 | Phase 5 | Complete |
| EVL-01 | Phase 6 | Complete |
| EVL-02 | Phase 6 | Complete |
| EVL-03 | Phase 6 | Complete |
| EVL-04 | Phase 6 | Complete |
| EVL-05 | Phase 6 | Complete |
| EVL-06 | Phase 6 | Complete |
| EVL... | adapter-lab | .planning/milestones/v1.0-REQUIREMENTS.md | Markdown | 51a6754e11ed930621fe64657388f7d16cb88106e89c2ab156a551f2481eb230 | 6 | 587 |
# Roadmap: adapter-lab
## Overview
The v1 roadmap builds `adapter-lab` as an enterprise-grade vertical chain: establish runtime contracts, manifests, artifact storage, plugin trust, run-state, and offline policy first; then deliver ingest, synthesis, training, evaluation, verification, and signed bundle production as... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | bf11582b3cab68f0ce4f63c0fc230bed08397a11c7272e20f0468cee3575d074 | 0 | 896 |
Wave 1 completion)*
- [x] 01-02: Config, stage manifest, run manifest, bundle manifest, and JSON Schema export models
**Wave 3** *(blocked on Wave 2 completion)*
- [x] 01-03: Content-addressed artifact store, immutable blob refs, run-state, and stage DAG contracts
- [x] 01-04: Plugin registry, entry-point groups, al... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | 786f4dae35769ac06ef2746d77f7710f697f27c58504ea8e81d7a551813bb3d9 | 1 | 896 |
** mvp
**Depends on:** Phase 2
**Requirements:** ING-01, ING-02, ING-03, ING-04, ING-05, ING-06, ING-07, ING-08, ING-09, ING-10, ING-11, ING-12, OPS-09
**Success Criteria** (what must be TRUE):
1. Operator can ingest representative PDF, DOCX, HTML, EML, Markdown, text, CSV, Parquet, and JSONL fixtures from local and... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | 6726e7fb5dc94b1aec7d7c8be7780f43b759455066dee20536a93cc0ba83d634 | 2 | 896 |
Wave 3** *(blocked on Wave 2 completion)*
- [x] 04-03: Preference-pair generation, teacher/judge model resolution, egress-policy integration, and quality filter pipeline
**Wave 4** *(blocked on Wave 3 completion)*
- [x] 04-04: Canary handling, synth manifests, deterministic tests, and privacy-safe report/bundle guar... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | 7a6207c1f9205534434fb679b80340b9655fd9943d8605af142753cf603d9b20 | 3 | 896 |
01: Evaluator protocol, lm-evaluation-harness integration, dataset refs, grader refs, and local resource policy
- [x] 06-02: Multilingual, instruction-following, BFCL/tau-bench, tool-inventory, and safety suite plugins
- [x] 06-03: Sanitized report bundle, base-vs-adapter comparison outputs, and evaluation manifest
- [... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | 4b350c2dacb266c8b515f1795a157c6dbff0b5a0073a3a5fc905694a1fd71b7b | 4 | 896 |
.
5. `verify-bundle` succeeds offline for intact bundles and fails on tampered adapter bytes, manifests, attestations, signatures, model cards, reports, dashboards, and reproducibility metadata.
6. Bundle reports, model cards, dashboards, and attestations never expose raw PII values, canary strings, signing secrets... | adapter-lab | .planning/milestones/v1.0-ROADMAP.md | Markdown | 9475347e8eba381c4fedf4082bbe14a8ec7ed69e18df733304475ae749509c0a | 5 | 533 |
---
phase: 01-core-runtime-and-contracts
plan: 01-01
title: Package skeleton, CLI shell, SDK service boundaries, JSON output, and exit-code contract
type: execute
wave: 1
depends_on: []
files_modified:
- pyproject.toml
- uv.lock
- README.md
- src/adapterlab/__init__.py
- src/adapterlab/cli.py
- src/adapterl... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-01-PLAN.md | Markdown | da48bb37da9c653b2945b55ed06f863b5130cd303210cd76c87a62eb488e4fe0 | 0 | 896 |
, `ruff`. Create `src/adapterlab/__init__.py`, `src/adapterlab/sdk/__init__.py`, `tests/conftest.py`, and lock dependencies with `uv lock`.</action>
<acceptance_criteria>
- `pyproject.toml` contains `[project.scripts]` with `adapterlab = "adapterlab.cli:app"`.
- `src/adapterlab/__init__.py` defines `__v... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-01-PLAN.md | Markdown | 60a2c754d2c24e0d2c1b17deb0d1491110cebc6678d9f49b1be236caa9ba1c39 | 1 | 896 |
-q</verify>
</task>
<task id="01-01-04" type="docs">
<name>Document foundation CLI contract</name>
<read_first>
- `README.md`
- `vision.md` section 10
</read_first>
<action>Create or update `README.md` with a concise "Foundation CLI contract" section listing `--json`, stdout/stderr beha... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-01-PLAN.md | Markdown | 2619f7caad360f2a05d790626360bf1cd34288df03658c4ae0521a4c365cb383 | 2 | 307 |
---
phase: 01-core-runtime-and-contracts
plan: 01-01
subsystem: foundation
tags: [python, uv, typer, pydantic, pytest, ruff, cli, sdk]
requires: []
provides:
- Installable Python package skeleton with adapterlab import package
- Typer CLI shell with strict JSON stdout contract
- SDK error and output primitives wi... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-01-SUMMARY.md | Markdown | ebba3f0c45d6501cb03e19dbbf20c07d2d3e19afe0ff8659ebcafcedab8a0f1c | 0 | 896 |
.py` - Import package version.
- `src/adapterlab/cli.py` - Typer CLI shell and foundation command groups.
- `src/adapterlab/sdk/__init__.py` - SDK namespace.
- `src/adapterlab/sdk/errors.py` - Stable SDK exception classes and exit-code mapping.
- `src/adapterlab/sdk/output.py` - Typed JSON command result envelope.
- `t... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-01-SUMMARY.md | Markdown | bcc5eba5c7efb0df922b36129c1c8122229af48970d27d4b5cdb053d6e862ab7 | 1 | 583 |
---
phase: 01-core-runtime-and-contracts
plan: 01-02
title: Config, stage manifest, run manifest, bundle manifest, and JSON Schema export models
type: tdd
wave: 2
depends_on: [01-01]
files_modified:
- src/adapterlab/contracts/__init__.py
- src/adapterlab/contracts/base.py
- src/adapterlab/contracts/config.py
- ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-02-PLAN.md | Markdown | b7ae45abb7a0e2bf32a9301e3904a32b374cae6fe435750ca8647d3cdf613a06 | 0 | 896 |
mode, artifact ref, stage manifest, run config, run manifest, bundle manifest, and plugin record. Create `tests/contracts/test_schema_exports.py` asserting schema files exist under `schemas/` and generated schemas match committed files.</action>
<acceptance_criteria>
- Tests include invalid digest format, unk... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-02-PLAN.md | Markdown | 491f50e3fd137463952e5e0a584c642b43caf3d207efe4887d0657d5c9058179 | 1 | 896 |
contracts import TenantConfig, RunManifest, BundleManifest"` exits 0.
- `README.md` contains `adapterlab schemas validate`.
- `uv run pytest tests/contracts tests/cli -q` exits 0.
</acceptance_criteria>
<verify>uv run pytest tests/contracts tests/cli -q</verify>
</task>
</tasks>
<verification>
- ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-02-PLAN.md | Markdown | 48d323b7a10072c3d44ee63c5c07586d67a8ab433d7fc6aaa297eb2a0f8748a0 | 2 | 143 |
---
phase: 01-core-runtime-and-contracts
plan: 01-02
subsystem: contracts
tags: [pydantic, json-schema, yaml, manifests, config, cli, pytest]
requires:
- phase: 01-core-runtime-and-contracts
provides: Package skeleton, CLI shell, SDK error/output primitives
provides:
- Strict Pydantic foundation contract models... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-02-SUMMARY.md | Markdown | 60c1a00b783d394f1713ff3b9802858987c62a28da04c8f3a978fc4c8f843d74 | 0 | 896 |
** - `4d7e4c8` (`feat`)
4. **Task 01-02-04: Consolidate contract exports and documentation** - `5edce77` (`docs`)
**Plan metadata:** this summary commit.
## Files Created/Modified
- `src/adapterlab/contracts/base.py` - Strict versioned contract base model.
- `src/adapterlab/contracts/config.py` - Deployment mode and... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-02-SUMMARY.md | Markdown | c2bc0367684a17a2594c615577eafc6701d5970981eeca50c85a819160f87307 | 1 | 686 |
---
phase: 01-core-runtime-and-contracts
plan: 01-03
title: Content-addressed artifact store, immutable blob refs, run-state, and stage DAG contracts
type: tdd
wave: 3
depends_on: [01-01, 01-02]
files_modified:
- src/adapterlab/artifacts/__init__.py
- src/adapterlab/artifacts/store.py
- src/adapterlab/artifacts/m... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-03-PLAN.md | Markdown | 1edb6b86288e31b729e252efdc262fd67d9fca1f206afcdf9c75a609722d50f2 | 0 | 896 |
` contains tests for cycles, duplicate stage IDs, and unknown dependencies.
- `tests/run/test_stage_state.py` contains tests for legal and illegal stage-state transitions.
- `uv run pytest tests/artifacts tests/run -q` fails before implementation because modules are missing.
</acceptance_criteria>
<... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-03-PLAN.md | Markdown | 0ef9ee3a4a2f7f94806695fbe2da52c63ce041f651191ba0ec07f91e2a992b23 | 1 | 842 |
---
phase: 01-core-runtime-and-contracts
plan: 01-03
subsystem: artifacts-run
tags: [cas, sha256, artifacts, run-dag, stage-state, typer, pytest]
requires:
- phase: 01-core-runtime-and-contracts
provides: Foundation contracts and schema validation
provides:
- Immutable local content-addressed artifact store
-... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-03-SUMMARY.md | Markdown | 5281741136f575a48a76204d861d577176af1c16799c14d4891e08a5b875311c | 0 | 896 |
sidecar payload helper.
- `src/adapterlab/run/dag.py` - Topological dry-run validation and resume pointer checks.
- `src/adapterlab/run/state.py` - Stage-state transition validation.
- `src/adapterlab/cli.py` - `artifacts check` and `run-dag validate` commands.
- `tests/artifacts/*` - CAS behavior and tamper tests.
- `... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-03-SUMMARY.md | Markdown | e217e025a0f43ee76f0f9b696227a5805af2ca8f2fdd8b4d7616b1e1fcfb9e85 | 1 | 372 |
---
phase: 01-core-runtime-and-contracts
plan: 01-04
title: Plugin registry, entry-point groups, allowlist rules, API-version checks, and package hash evidence
type: tdd
wave: 3
depends_on: [01-01, 01-02]
files_modified:
- pyproject.toml
- src/adapterlab/plugins/__init__.py
- src/adapterlab/plugins/registry.py
... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-04-PLAN.md | Markdown | 93b119a56dcde1699b2aace10f0a1e39fb1b561d0634061256a484f7fa3159d1 | 0 | 896 |
green">
<name>Implement entry-point discovery and plugin records</name>
<read_first>
- `tests/plugins/test_registry.py`
- `src/adapterlab/contracts/plugins.py`
</read_first>
<action>Implement plugin discovery over known groups `adapterlab.loaders`, `adapterlab.pii_recognizers`, `adapterlab.l... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-04-PLAN.md | Markdown | 044cf78437e7d2391e4b89f8018267eaec7b74633a8292b3180572a1b1596439 | 1 | 737 |
---
phase: 01-core-runtime-and-contracts
plan: 01-04
subsystem: plugins
tags: [plugins, entry-points, allowlist, package-evidence, cli, pytest]
requires:
- phase: 01-core-runtime-and-contracts
provides: Foundation contracts and CLI shell
provides:
- Entry-point-only plugin discovery for known adapterlab groups
... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-04-SUMMARY.md | Markdown | 8390260b36d58c6a6256bd8e40870c3c5081e7eb17e67f98f6c2a24e1dacdbad | 0 | 896 |
record object reference field.
- `schemas/plugin-record.schema.json` - Updated plugin schema export.
- `src/adapterlab/cli.py` - `plugins list` command.
- `tests/plugins/*` - Plugin discovery, allowlist, and evidence tests.
- `tests/cli/test_plugins_cli.py` - Plugin listing CLI JSON test.
## Decisions Made
- Treated ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-04-SUMMARY.md | Markdown | 6ca9ef1da3a93e34b7b560e976b42a43c4262613a85af82813d79d652ceaa136 | 1 | 276 |
---
phase: 01-core-runtime-and-contracts
plan: 01-05
title: Initialization workflow, default policy templates, tenant config templates, and schema validation tests
type: tdd
wave: 4
depends_on: [01-02, 01-03, 01-04]
files_modified:
- src/adapterlab/init/__init__.py
- src/adapterlab/init/scaffold.py
- src/adapterl... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-05-PLAN.md | Markdown | 59c350bf93bb3d6341b3ea68dd7d7344b01a4b05f3383fcd055667b924535658 | 0 | 896 |
airgapped and empty allowlists. |
| T-05-INIT-OVERWRITE | medium | Scaffolder refuses overwrites unless explicit force is passed and reports each file action. |
| T-05-INIT-PLUGIN-RCE | high | Scaffold writes plugin allowlist templates only, never executable plugin code in corpus/source paths. |
</threat_model>
<tasks... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-05-PLAN.md | Markdown | c63a24c60e79ef73d15cd8700a6437ef0f49ce3bf02f164c12cf3dd53e1919c0 | 1 | 896 |
0` and prints parseable JSON.
- The JSON output includes `target`, `created`, `skipped`, `changed`, `deployment_mode`, and `artifact_store`.
- Existing target files without force produce a config or validation error with a nonzero documented exit code and no partial overwrite.
- `uv run pytest tests/i... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-05-PLAN.md | Markdown | d38d3d573d1461dfe0d4368e605ea12ca6e6ff8cfb4ee035516cc8c97db2f5c8 | 2 | 657 |
---
phase: 01-core-runtime-and-contracts
plan: 01-05
subsystem: init
tags: [init, scaffold, templates, airgapped, privacy, cli, pytest]
requires:
- phase: 01-core-runtime-and-contracts
provides: Foundation schemas, artifact store, run DAG, and plugin trust services
provides:
- Safe airgapped tenant workspace in... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-05-SUMMARY.md | Markdown | 46a81017e8cbc7acb4fe7a0137a54f3931d45e201274fc3cc10ec7ce8cd032c0 | 0 | 896 |
` (`feat`)
4. **Task 01-05-04: Add privacy and no-network-safe scaffold checks** - `f9a0c6a` (`feat`)
5. **Task 01-05-05: Document Phase 1 foundation workflow** - `eecff8d` (`docs`)
**Plan metadata:** this summary commit.
## Files Created/Modified
- `src/adapterlab/init/scaffold.py` - SDK scaffolder and file action ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-05-SUMMARY.md | Markdown | 53852bef5194bf0f61ba0480108f6966095f693ed88f958df73f57c51215b975 | 1 | 414 |
# Phase 1: Core Runtime and Contracts - Context
**Gathered:** 2026-05-26
**Status:** Ready for planning
<domain>
## Phase Boundary
Phase 1 delivers the foundation that every later stage depends on: the Python package skeleton, Typer CLI shell, importable SDK service boundaries, versioned config and manifest models, ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-CONTEXT.md | Markdown | e16e0571fafe628581463405fa5a6495aec8c99a326903d63b939622105bf474 | 0 | 896 |
/digest evidence.
- **D-15:** Plugin API-version compatibility fails closed. Missing or incompatible plugin API versions are validation errors and cannot be activated.
- **D-16:** Plugin listing reports both trusted and untrusted discovered plugins with reasons, but runtime use and manifest inclusion refuse unallowlist... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-CONTEXT.md | Markdown | 43735e19318655ad6ee24a54afbff3d30aa4a4f4b1c481d072bf5c8599dc9298 | 1 | 896 |
`, `peft`, `datasets`, and `trl`.
- Treat `adapterlab init` as an enterprise-safe scaffold, not a demo project generator.
- Treat schemas and exit codes as public contracts from the first implementation phase.
</specifics>
<deferred>
## Deferred Ideas
None — discussion stayed within Phase 1 scope.
</deferred>
---
... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-CONTEXT.md | Markdown | fa91c670e0a8f52343d3fd232b22093f6039ad2548a61934d7c51728c0144707 | 2 | 96 |
# Phase 1: Core Runtime and Contracts - Discussion Log
> **Audit trail only.** Do not use as input to planning, research, or execution agents.
> Decisions are captured in CONTEXT.md — this log preserves the alternatives considered.
**Date:** 2026-05-26
**Phase:** 1-Core Runtime and Contracts
**Areas discussed:** CLI ... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-DISCUSSION-LOG.md | Markdown | 6bcaa30684f2eb459ef35565428705785f5eacedc461a8c171c9ef615a991bcd | 0 | 896 |
----|-------------|----------|
| Strict structural validation | Required fields, versions, digests, enums, stage IDs, and refs are enforced now. | x |
| Strict only for Phase 1-owned fields | Validate basics but allow future payloads as flexible objects. | |
| Permissive early | Prioritize schema generation and round-t... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-DISCUSSION-LOG.md | Markdown | c90df6087f734cfe0b6bd8330b72b5db3c00fe7c2b50c9e417a04eabcd6c3a43 | 1 | 896 |
boundary | Allow rich Python objects internally, serialize at command boundaries. | |
| Flexible early | Enforce manifest-only handoffs after subsystem implementation. | |
**User's choice:** Manifest-only handoff APIs.
**Notes:** Protects the central manifest architecture from shortcut coupling.
---
## Plugin Discov... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-DISCUSSION-LOG.md | Markdown | a47b0e94e7154228df12402cef066cdfae701d9d5729831b6c2bb6fdffd68723 | 2 | 896 |
** Airgapped by default.
**Notes:** Aligns with offline-first product constraints.
| Option | Description | Selected |
|--------|-------------|----------|
| Safe and idempotent | Refuse overwrite by default; support explicit force; report created/skipped/changed files. | x |
| Overwrite templates | Replace existing ge... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-DISCUSSION-LOG.md | Markdown | a98eed90b425d5462fd6727692698b94735087080632f3131d05a5f09bfc0b34 | 3 | 299 |
# Phase 01 - Pattern Map
## Codebase State
No application/package source code exists yet. The closest reusable assets are planning artifacts:
- `.planning/PROJECT.md` for product boundary and accepted decisions.
- `.planning/REQUIREMENTS.md` for FND requirement IDs.
- `.planning/ROADMAP.md` for the five Phase 1 plan... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-PATTERNS.md | Markdown | 508fd5f47b9bcb37632918d10a056073b3b72e7edaba0160fe2a580c9dbe26cd | 0 | 403 |
# Phase 1: Core Runtime and Contracts - Research
## RESEARCH COMPLETE
**Question answered:** What needs to be known to plan Phase 1 well?
## Scope Summary
Phase 1 is a foundation phase for a Python package, not a model pipeline implementation. The first usable capability is: an operator can install/import `adapterl... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-RESEARCH.md | Markdown | 09e65e49c16a697108fe2e34593b7c657c7814d1ab7ccd6f6d0478f4ff2b849e | 0 | 896 |
as artifact identity; the digest ref is the durable contract.
- Do not use arbitrary dicts for future manifests where strict envelopes and extension points can keep contracts stable.
## Source Notes
- `.planning/STACK.md` selects Python, `uv`, Typer, Pydantic v2, committed schemas, and pytest/Hypothesis.
- `vision.md... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-RESEARCH.md | Markdown | 72f01a4ff6b406ed282626533835a1ba186702d2afddd835cab936d77d46f9e0 | 1 | 131 |
---
phase: 01
slug: core-runtime-and-contracts
status: verified
threats_open: 0
asvs_level: 1
created: 2026-05-26T19:45:25Z
updated: 2026-05-26T19:45:25Z
---
# Phase 01 — Security
Per-phase security contract for Phase 01: Core Runtime and Contracts.
## Trust Boundaries
| Boundary | Description | Data Crossing |
|--... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-SECURITY.md | Markdown | 5299823453c35d8aef63d932916cd8d024f55b08192a9810f96e991aade47b8e | 0 | 896 |
and ignores filesystem/path-style plugin sources. | closed |
| T-04-PLUGIN-IMPERSONATION | Spoofing | Plugin allowlist | mitigate | Trust requires matching name, entry-point group, API version, package name, package version, and package digest. | closed |
| T-04-PLUGIN-API-DRIFT | Tampering | Plugin activation | mitiga... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-SECURITY.md | Markdown | 434077929e71b84cabda021e048b3dae2c8616743593f4c72c16570f95f8a1fd | 1 | 536 |
# Walking Skeleton - adapter-lab
**Phase:** 1
**Generated:** 2026-05-26
## Capability Proven End-to-End
An operator can install/import `adapterlab`, run `adapterlab init --json` for an airgapped tenant workspace, validate generated configs/schemas, write and inspect a content-addressed artifact, validate a dry-run r... | adapter-lab | .planning/phases/01-core-runtime-and-contracts/01-SKELETON.md | Markdown | 610e030da22b86a0fe287670c47ce27960c02ed60511154fa7bb2106f4a47ce7 | 0 | 555 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.