sindri-forge 0.3.2__tar.gz

sindri_forge-0.3.2/.claude-plugin/marketplace.json

@@ -0,0 +1,33 @@
+{
+  "name": "4kmetrics",
+  "owner": {
+    "name": "4KMetrics",
+    "email": "260103116+4kmetrics@users.noreply.github.com"
+  },
+  "metadata": {
+    "description": "Community plugins from 4KMetrics"
+  },
+  "plugins": [
+    {
+      "name": "sindri",
+      "source": "./",
+      "description": "Bounded, target-driven optimization loops for any metric you can measure deterministically. Autonomous experiments, kept-or-reverted per git commit, auto-PR on success.",
+      "author": {
+        "name": "Nimesh Kumar",
+        "email": "260103116+4kmetrics@users.noreply.github.com"
+      },
+      "homepage": "https://github.com/4KMetrics/sindri",
+      "repository": "https://github.com/4KMetrics/sindri",
+      "license": "MIT",
+      "keywords": [
+        "optimization",
+        "benchmarking",
+        "autonomous",
+        "experiments",
+        "performance",
+        "ci",
+        "bundle-size"
+      ]
+    }
+  ]
+}

sindri_forge-0.3.2/.claude-plugin/plugin.json

@@ -0,0 +1,21 @@
+{
+  "name": "sindri",
+  "description": "Bounded, target-driven optimization loops for any metric you can measure deterministically. Autonomous experiments, kept-or-reverted per git commit, auto-PR on success.",
+  "version": "0.3.2",
+  "author": {
+    "name": "Nimesh Kumar",
+    "email": "260103116+4kmetrics@users.noreply.github.com"
+  },
+  "homepage": "https://github.com/4KMetrics/sindri",
+  "repository": "https://github.com/4KMetrics/sindri",
+  "license": "MIT",
+  "keywords": [
+    "optimization",
+    "benchmarking",
+    "autonomous",
+    "experiments",
+    "performance",
+    "ci",
+    "bundle-size"
+  ]
+}

sindri_forge-0.3.2/.github/workflows/ci.yml

@@ -0,0 +1,48 @@
+name: ci
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    name: pytest (${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+
+      - name: Configure git (required for git_ops + e2e)
+        run: |
+          git config --global user.email "ci@sindri.local"
+          git config --global user.name "Sindri CI"
+
+      - name: Install
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ".[dev]"
+
+      - name: Fast tests (unit + integration, skip e2e)
+        run: pytest -m "not e2e" -q
+
+      - name: End-to-end tests
+        if: matrix.python-version == '3.12'
+        run: pytest -m e2e -q
+
+      - name: Smoke
+        if: matrix.python-version == '3.12'
+        run: ./scripts/smoke.sh

sindri_forge-0.3.2/.github/workflows/publish.yml

@@ -0,0 +1,68 @@
+name: publish
+
+on:
+  workflow_run:
+    workflows: ["release-please"]
+    types: [completed]
+  workflow_dispatch:
+
+permissions:
+  id-token: write  # required for PyPI trusted publisher (OIDC)
+  contents: read
+
+jobs:
+  pypi:
+    name: build + publish to PyPI
+    runs-on: ubuntu-latest
+    # workflow_run fires on every release-please completion (PR updates + real
+    # releases). Only proceed when the upstream run succeeded, and then let the
+    # release-tag check inside the job decide whether a real release happened.
+    if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          # Fetch full history + tags so we can detect release commits.
+          fetch-depth: 0
+          ref: ${{ github.event.workflow_run.head_sha || github.sha }}
+
+      - name: Detect release tag at HEAD
+        id: gate
+        run: |
+          set -euo pipefail
+          git fetch --tags --force
+          # HEAD is a release when the current commit has a matching v<semver> tag.
+          if TAG=$(git describe --tags --exact-match HEAD 2>/dev/null); then
+            echo "tag=$TAG" >> "$GITHUB_OUTPUT"
+            echo "is_release=true" >> "$GITHUB_OUTPUT"
+            echo ">> HEAD is release tag: $TAG"
+          else
+            echo "is_release=false" >> "$GITHUB_OUTPUT"
+            echo ">> HEAD has no exact release tag — skipping publish."
+          fi
+
+      - name: Set up Python
+        if: steps.gate.outputs.is_release == 'true'
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install build tooling
+        if: steps.gate.outputs.is_release == 'true'
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build
+
+      - name: Build wheel + sdist
+        if: steps.gate.outputs.is_release == 'true'
+        run: python -m build
+
+      - name: Publish to PyPI
+        if: steps.gate.outputs.is_release == 'true'
+        uses: pypa/gh-action-pypi-publish@release/v1
+        # Uses PyPI trusted publishing — no API token needed.
+        # One-time setup on PyPI (https://pypi.org/manage/account/publishing/):
+        #   PyPI Project Name: sindri-forge
+        #   Owner: 4KMetrics
+        #   Repository name: sindri
+        #   Workflow name: publish.yml
+        #   Environment name: (blank)

sindri_forge-0.3.2/.github/workflows/release-please.yml

@@ -0,0 +1,19 @@
+name: release-please
+
+on:
+  push:
+    branches: [main]
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  release-please:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: googleapis/release-please-action@v4
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          config-file: release-please-config.json
+          manifest-file: .release-please-manifest.json

sindri_forge-0.3.2/.gitignore

@@ -0,0 +1,33 @@
+# sindri runtime state (always local, never committed)
+.sindri/
+
+# Claude Code / superpowers scratch
+.superpowers/
+.remember/
+
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+*.egg-info/
+dist/
+build/
+.venv/
+venv/
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Editors
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# Environment
+.env
+.env.local

sindri_forge-0.3.2/.markdownlint.json

@@ -0,0 +1,8 @@
+{
+  "default": true,
+  "MD013": false,
+  "MD032": false,
+  "MD033": false,
+  "MD041": false,
+  "MD060": false
+}

sindri_forge-0.3.2/.python-version

@@ -0,0 +1 @@
+3.10

sindri_forge-0.3.2/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.3.2"
+}

sindri_forge-0.3.2/CHANGELOG.md

@@ -0,0 +1,82 @@
+# Changelog
+
+## [0.3.2](https://github.com/4KMetrics/sindri/compare/v0.3.1...v0.3.2) (2026-04-21)
+
+
+### Bug Fixes
+
+* version from pyproject metadata; publish via workflow_run ([#12](https://github.com/4KMetrics/sindri/issues/12)) ([de7a1aa](https://github.com/4KMetrics/sindri/commit/de7a1aafda88692f75c2da73981cf1624461e639))
+
+## [0.3.1](https://github.com/4KMetrics/sindri/compare/v0.3.0...v0.3.1) (2026-04-21)
+
+
+### Bug Fixes
+
+* **ci:** rename PyPI dist to sindri-forge; publish on tag push ([#10](https://github.com/4KMetrics/sindri/issues/10)) ([432b50c](https://github.com/4KMetrics/sindri/commit/432b50c4247fae56bcaf17e8c5a0096cd8812871))
+
+## [0.3.0](https://github.com/4KMetrics/sindri/compare/v0.2.0...v0.3.0) (2026-04-21)
+
+
+### Features
+
+* **plugin:** add community marketplace manifest + metadata enrichment ([#7](https://github.com/4KMetrics/sindri/issues/7)) ([fff07c5](https://github.com/4KMetrics/sindri/commit/fff07c5dea056041c38011744ba8126e51c816fb))
+
+## [0.2.0](https://github.com/4KMetrics/sindri/compare/v0.1.1...v0.2.0) (2026-04-21)
+
+
+### Features
+
+* user-ready shipping (install script + dogfood + PyPI + pre-flight) ([#6](https://github.com/4KMetrics/sindri/issues/6)) ([37066ed](https://github.com/4KMetrics/sindri/commit/37066ed850e824149161d4ab10b0fa064b6567aa))
+
+
+### Bug Fixes
+
+* **ci:** release-please feat commits should bump minor pre-1.0 ([#4](https://github.com/4KMetrics/sindri/issues/4)) ([db4fe94](https://github.com/4KMetrics/sindri/commit/db4fe94426353f81c9c4e2dba9587f1ad38be594))
+
+## [0.1.1](https://github.com/4KMetrics/sindri/compare/v0.1.0...v0.1.1) (2026-04-20)
+
+
+### Features
+
+* **cli:** add archive subcommand ([4d00181](https://github.com/4KMetrics/sindri/commit/4d001811da7d7f281161a28fc70ead9074b42651))
+* **cli:** add check-termination subcommand ([c93ccdb](https://github.com/4KMetrics/sindri/commit/c93ccdb673ae2c7b52ac9408339dcec6ba211529))
+* **cli:** add detect-mode subcommand ([d596cb3](https://github.com/4KMetrics/sindri/commit/d596cb3f6b84c67d867cd9f9242d6e90ff597d06))
+* **cli:** add generate-pr-body subcommand ([e5e629c](https://github.com/4KMetrics/sindri/commit/e5e629cdcc139162e36f08a0541fe2faf85e1830))
+* **cli:** add init subcommand for end-to-end run setup ([4b766f7](https://github.com/4KMetrics/sindri/commit/4b766f723b538c95c3bfe9419690785c95b76ad1))
+* **cli:** add pick-next subcommand ([03c21bd](https://github.com/4KMetrics/sindri/commit/03c21bdbf53514cfbc7599039b0b03cf83585afe))
+* **cli:** add read-state subcommand ([6b81247](https://github.com/4KMetrics/sindri/commit/6b81247d7e6b2b9956eaaa9279a7c66fb912f6cd))
+* **cli:** add record-result subcommand ([11c3c39](https://github.com/4KMetrics/sindri/commit/11c3c39b6ab0e95c7867ed4ef8f980487adc62fc))
+* **cli:** add status subcommand ([48dd080](https://github.com/4KMetrics/sindri/commit/48dd08057af680c88ff9ac309de4ef45e9519c13))
+* **cli:** add validate-benchmark subcommand ([5a4eb0f](https://github.com/4KMetrics/sindri/commit/5a4eb0f01e1ecd1f7272e085291a1dc505e71491))
+* **cli:** scaffold argparse dispatcher with subcommand registry ([f49b82f](https://github.com/4KMetrics/sindri/commit/f49b82f44135ef1331277c1fc18135d8033ef09d))
+* **core:** add git subprocess wrappers ([444abe6](https://github.com/4KMetrics/sindri/commit/444abe65db19fcb6854ea3854f83011ab59ad82d))
+* **core:** add local/remote mode detection ([7fa9710](https://github.com/4KMetrics/sindri/commit/7fa971016022543713ea4c06567b4dce6d6c9068))
+* **core:** add METRIC line parser ([a0a325f](https://github.com/4KMetrics/sindri/commit/a0a325fcedd3e7fc9982a415ca172d51a9ba14a8))
+* **core:** add noise floor, confidence ratio, CV stats ([fb2ddea](https://github.com/4KMetrics/sindri/commit/fb2ddeab446470cb1d0c62b3cfdbc4937d4bee6b))
+* **core:** add pool ordering and termination predicates ([c96168a](https://github.com/4KMetrics/sindri/commit/c96168ab0aa8f91d3efa4ecb7aaf75eeeb833928))
+* **core:** add PR body markdown renderer ([b532b3d](https://github.com/4KMetrics/sindri/commit/b532b3dc86f174ed8ca1e6c55238adf29d363e2c))
+* **core:** add pydantic models for state, jsonl, subagent result ([e5d1308](https://github.com/4KMetrics/sindri/commit/e5d1308abe8cf80aadf27211348bbd5f7df3c437))
+* **core:** add state.py for sindri.md and sindri.jsonl I/O ([bc3131d](https://github.com/4KMetrics/sindri/commit/bc3131d643447bd26ee36ce24351d46fab32be91))
+* **core:** add termination decision table composing predicates ([6fc4f83](https://github.com/4KMetrics/sindri/commit/6fc4f830b0c93ace249d2cb508679142dc7ca96a))
+* initial drop — design spec + Python core + CLI (Plan 1) ([10db076](https://github.com/4KMetrics/sindri/commit/10db076fd4d4026b3832b36dfae6cefe3f0b713d))
+* **plugin:** sindri plugin artifacts (Plan 2) ([#2](https://github.com/4KMetrics/sindri/issues/2)) ([62cdb0f](https://github.com/4KMetrics/sindri/commit/62cdb0fa59f1aeeb558b8a456a62cd5250b55934))
+
+
+### Bug Fixes
+
+* **cli:** init fails fast on taken branch; goal regex uses fullmatch ([c92761f](https://github.com/4KMetrics/sindri/commit/c92761f7962e723af15f6755d649fe13f945d5af))
+* **cli:** refuse unsafe slugs when archiving ([3b9b4e6](https://github.com/4KMetrics/sindri/commit/3b9b4e6d59b9f8de315dbbf94daf7e6665ff2451))
+* code-review findings (StateIOError guards + wget localhost parity) ([6bd7257](https://github.com/4KMetrics/sindri/commit/6bd72578cf0a5c10a2cf390c9bbf681dbb8dce7c))
+* **state:** surface malformed jsonl skips on stderr; clarify atomicity comment ([660b143](https://github.com/4KMetrics/sindri/commit/660b1432f44a63276584bfa546b9867a45db3054))
+* **validators:** tz-aware started_at; None-sentinel mode-based defaults ([480ede3](https://github.com/4KMetrics/sindri/commit/480ede3e68246d3ca3969962432ca17671e0d24d))
+
+
+### Documentation
+
+* add initial design spec for sindri v1 ([f797819](https://github.com/4KMetrics/sindri/commit/f7978195ea9a0cff15aff22231d2aa951b133034))
+* add Plan 1 — sindri Python core + CLI implementation ([6be79c6](https://github.com/4KMetrics/sindri/commit/6be79c603c66cd2093b75bc8a048b32aabbf3dcb))
+* add Table of Contents to sindri design spec ([fcf2af9](https://github.com/4KMetrics/sindri/commit/fcf2af9f4b17b30a9da0af14403355219d21c6ec))
+* clarify benchmark scaffolding is sindri's job, not a separate chat ([c4eadd5](https://github.com/4KMetrics/sindri/commit/c4eadd57f670a3342f62cd5941fd37457e610614))
+* drop personal-roadmap framing from sindri spec ([f8ba7df](https://github.com/4KMetrics/sindri/commit/f8ba7dfd712e75b09f859d900e9267f6775cf7bb))
+* replace ratchet metaphor with blacksmith-native language ([f5cc5b1](https://github.com/4KMetrics/sindri/commit/f5cc5b1d50349d1fd3eda4379f304fcac4be4b0b))
+* switch user benchmark scripts from shell to Python ([eb647d0](https://github.com/4KMetrics/sindri/commit/eb647d0df634f6c1ba2ad3a0deb8dc4f2620fbf1))

sindri_forge-0.3.2/PKG-INFO

@@ -0,0 +1,110 @@
+Metadata-Version: 2.4
+Name: sindri-forge
+Version: 0.3.2
+Summary: Claude Code plugin core: bounded, target-driven optimization loops
+Author: Nimesh Kumar
+License: MIT
+Requires-Python: >=3.10
+Requires-Dist: pydantic<3.0,>=2.0
+Provides-Extra: dev
+Requires-Dist: pytest-cov>=4.1; extra == 'dev'
+Requires-Dist: pytest-mock>=3.12; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: pyyaml>=6.0; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# sindri
+
+A Claude Code plugin for bounded, target-driven optimization loops.
+
+> In the myth, Sindri is the dwarf smith who forged Mjölnir by iterating under adversity — each hammer-strike tested, kept only if it survived. This plugin is the same pattern: make a change, benchmark it, keep it if it's better, revert otherwise, repeat until the target is hit or the pool of ideas is exhausted.
+
+Inspired by [karpathy/autoresearch](https://github.com/karpathy/autoresearch) and [davebcn87/pi-autoresearch](https://github.com/davebcn87/pi-autoresearch), adapted to Claude Code.
+
+**One commit per kept experiment. Git is the audit log. The forge runs one way.**
+
+## Install
+
+### Option 1 — Claude Code marketplace (easiest, once PyPI is up)
+
+```
+/plugin marketplace add 4KMetrics/sindri
+/plugin install sindri@4kmetrics
+```
+
+Then install the Python runtime once (the plugin calls `python -m sindri`):
+
+```bash
+uv pip install sindri-forge                                      # or: pip install sindri-forge
+```
+
+The PyPI distribution is `sindri-forge` (the `sindri` name was taken by an unrelated ZK-proof SDK). The imported module name is still `sindri`.
+
+Restart Claude Code. Done.
+
+### Option 2 — local dev install (no marketplace)
+
+```bash
+git clone https://github.com/4KMetrics/sindri.git ~/src/sindri
+cd ~/src/sindri && ./scripts/install-plugin.sh
+```
+
+The install script handles the Python package + the `~/.claude/plugins/sindri` symlink in one go.
+
+That script installs the `sindri` Python package into your active environment and symlinks the plugin into `~/.claude/plugins/sindri`. Restart Claude Code afterward. Verify: `/sindri status` should report "no active run" cleanly.
+
+To target a specific Python (e.g. a venv), pass it via env var:
+
+```bash
+SINDRI_PYTHON=/path/to/venv/bin/python ./scripts/install-plugin.sh
+```
+
+Requirements:
+
+- Python ≥ 3.10 (pydantic 2 is the only runtime dep; pyyaml is dev-only)
+- `git` ≥ 2.30
+- `gh` CLI (authenticated) — needed only for `sindri-finalize` PR creation
+- Claude Code CLI with `Task` + `ScheduleWakeup` tools
+- `uv` (optional, faster install) or `pip`
+
+**First-run dogfood:** walk through [`docs/DOGFOOD.md`](docs/DOGFOOD.md) on a throwaway repo before pointing sindri at real code — it verifies every skill path and catches environment gaps before you commit an overnight run.
+
+## Quickstart
+
+```bash
+# In the repo you want to optimize:
+/sindri reduce bundle_bytes by 15%
+```
+
+Sindri will:
+
+1. Scaffold `.claude/scripts/sindri/benchmark.py` if missing (interactive; asks you to describe the measurement in natural language).
+2. Scan your repo and draft a pool of 10–30 candidates.
+3. Ask you to approve / edit the pool.
+4. Check out a new branch `sindri/reduce-bundle-bytes-15pct` and run the baseline 3×.
+5. Loop autonomously — one experiment per wakeup, ~60s between them.
+6. When the target hits (or the pool drains with ≥1 win), push the branch and open a PR.
+
+Check in anytime with `/sindri status`. Halt with `/sindri stop`. Abandon with `/sindri clear`.
+
+## Architecture
+
+- **Python core** (`src/sindri/`) — pure functions: statistics, state file I/O, git wrappers, pool ordering, termination predicates, PR body rendering.
+- **Skills** (`skills/sindri-*/SKILL.md`) — prose prompts for the decisions that need Claude's inference: scaffolding a benchmark, proposing a candidate pool, orchestrating the loop.
+- **Subagent prompt** (`prompts/experiment-subagent.md`) — the contract every experiment subagent follows. Fresh context per experiment, no context bleed.
+
+See [`docs/superpowers/specs/2026-04-19-sindri-design.md`](docs/superpowers/specs/2026-04-19-sindri-design.md) for the full design.
+
+## Development
+
+```bash
+uv pip install -e ".[dev]"
+pytest                                            # all test layers
+pytest -m "not e2e" -q                            # skip the slow end-to-end loop test
+pytest tests/unit/test_plugin_artifacts.py -v     # just artifact validity checks
+./scripts/smoke.sh                                # plumbing smoke in a throwaway repo
+```
+
+## License
+
+MIT

sindri_forge-0.3.2/README.md

@@ -0,0 +1,95 @@
+# sindri
+
+A Claude Code plugin for bounded, target-driven optimization loops.
+
+> In the myth, Sindri is the dwarf smith who forged Mjölnir by iterating under adversity — each hammer-strike tested, kept only if it survived. This plugin is the same pattern: make a change, benchmark it, keep it if it's better, revert otherwise, repeat until the target is hit or the pool of ideas is exhausted.
+
+Inspired by [karpathy/autoresearch](https://github.com/karpathy/autoresearch) and [davebcn87/pi-autoresearch](https://github.com/davebcn87/pi-autoresearch), adapted to Claude Code.
+
+**One commit per kept experiment. Git is the audit log. The forge runs one way.**
+
+## Install
+
+### Option 1 — Claude Code marketplace (easiest, once PyPI is up)
+
+```
+/plugin marketplace add 4KMetrics/sindri
+/plugin install sindri@4kmetrics
+```
+
+Then install the Python runtime once (the plugin calls `python -m sindri`):
+
+```bash
+uv pip install sindri-forge                                      # or: pip install sindri-forge
+```
+
+The PyPI distribution is `sindri-forge` (the `sindri` name was taken by an unrelated ZK-proof SDK). The imported module name is still `sindri`.
+
+Restart Claude Code. Done.
+
+### Option 2 — local dev install (no marketplace)
+
+```bash
+git clone https://github.com/4KMetrics/sindri.git ~/src/sindri
+cd ~/src/sindri && ./scripts/install-plugin.sh
+```
+
+The install script handles the Python package + the `~/.claude/plugins/sindri` symlink in one go.
+
+That script installs the `sindri` Python package into your active environment and symlinks the plugin into `~/.claude/plugins/sindri`. Restart Claude Code afterward. Verify: `/sindri status` should report "no active run" cleanly.
+
+To target a specific Python (e.g. a venv), pass it via env var:
+
+```bash
+SINDRI_PYTHON=/path/to/venv/bin/python ./scripts/install-plugin.sh
+```
+
+Requirements:
+
+- Python ≥ 3.10 (pydantic 2 is the only runtime dep; pyyaml is dev-only)
+- `git` ≥ 2.30
+- `gh` CLI (authenticated) — needed only for `sindri-finalize` PR creation
+- Claude Code CLI with `Task` + `ScheduleWakeup` tools
+- `uv` (optional, faster install) or `pip`
+
+**First-run dogfood:** walk through [`docs/DOGFOOD.md`](docs/DOGFOOD.md) on a throwaway repo before pointing sindri at real code — it verifies every skill path and catches environment gaps before you commit an overnight run.
+
+## Quickstart
+
+```bash
+# In the repo you want to optimize:
+/sindri reduce bundle_bytes by 15%
+```
+
+Sindri will:
+
+1. Scaffold `.claude/scripts/sindri/benchmark.py` if missing (interactive; asks you to describe the measurement in natural language).
+2. Scan your repo and draft a pool of 10–30 candidates.
+3. Ask you to approve / edit the pool.
+4. Check out a new branch `sindri/reduce-bundle-bytes-15pct` and run the baseline 3×.
+5. Loop autonomously — one experiment per wakeup, ~60s between them.
+6. When the target hits (or the pool drains with ≥1 win), push the branch and open a PR.
+
+Check in anytime with `/sindri status`. Halt with `/sindri stop`. Abandon with `/sindri clear`.
+
+## Architecture
+
+- **Python core** (`src/sindri/`) — pure functions: statistics, state file I/O, git wrappers, pool ordering, termination predicates, PR body rendering.
+- **Skills** (`skills/sindri-*/SKILL.md`) — prose prompts for the decisions that need Claude's inference: scaffolding a benchmark, proposing a candidate pool, orchestrating the loop.
+- **Subagent prompt** (`prompts/experiment-subagent.md`) — the contract every experiment subagent follows. Fresh context per experiment, no context bleed.
+
+See [`docs/superpowers/specs/2026-04-19-sindri-design.md`](docs/superpowers/specs/2026-04-19-sindri-design.md) for the full design.
+
+## Development
+
+```bash
+uv pip install -e ".[dev]"
+pytest                                            # all test layers
+pytest -m "not e2e" -q                            # skip the slow end-to-end loop test
+pytest tests/unit/test_plugin_artifacts.py -v     # just artifact validity checks
+./scripts/smoke.sh                                # plumbing smoke in a throwaway repo
+```
+
+## License
+
+MIT

sindri_forge-0.3.2/commands/sindri.md

@@ -0,0 +1,31 @@
+---
+description: Start or manage a sindri optimization run — bounded, target-driven experiments on any deterministic metric.
+argument-hint: <goal statement> | status | stop | scaffold-benchmark | clear
+---
+
+# /sindri — bounded optimization runs
+
+You are the front-of-house router for sindri. The user's input after `/sindri` is in `$ARGUMENTS`. Dispatch based on the first token:
+
+## Routing
+
+| First token of `$ARGUMENTS` | Action |
+|---|---|
+| `status` | Run `python -m sindri status` and print its output verbatim. No skill invocation. Zero-cost read. |
+| `stop` | `touch .sindri/current/HALT` (sentinel file — sindri-loop checks for it before every experiment and treats it as `halted_by_user` termination). Print `sindri: halt signal sent. The next wakeup will terminate without running another experiment.` |
+| `scaffold-benchmark` | Invoke skill `sindri-scaffold-benchmark` (no active run required). |
+| `clear` | Destructive. Read the branch name from `.sindri/current/sindri.md` JSON frontmatter **before** deleting anything. Prompt the user for confirmation (`y/N`). If confirmed: `rm -rf .sindri/current/ && git checkout main && git branch -D <branch>`. If not confirmed: print `aborted.` and stop. |
+| anything else | Treat as a goal statement. If `.sindri/current/` exists, print a resume/abandon prompt and **do not** start a new run. Otherwise, invoke skill `sindri-start` with the goal statement. |
+
+## Invariants
+
+- **One run per repo.** Never start a new run while `.sindri/current/` exists.
+- **Never mutate git state directly in this command.** Skills own that (except the `clear` branch delete, which is user-confirmed).
+- **`status` must always be fast.** It reads files, nothing else.
+- **`clear` always confirms first.** No silent deletion.
+
+## After dispatch
+
+For `status` and `stop`, you return to the user directly. For `scaffold-benchmark`, `clear`, and a new-goal invocation, control passes to the relevant skill and that skill produces the user-visible output.
+
+For a brand-new goal, after `sindri-start` returns, sindri is running autonomously — `ScheduleWakeup` has been set. The orchestrator skill (`sindri-loop`) runs on every wakeup without further user action until termination. On clean termination with ≥1 kept commit, the loop auto-invokes `sindri-finalize` to push the branch and open the PR.