text stringlengths 185 73.3k | repo stringlengths 7 100 | path stringlengths 4 146 | language stringclasses 7
values | hash stringlengths 16 16 | score float64 7 8.5 | stars int64 0 237k |
|---|---|---|---|---|---|---|
"""
qc.py - dashboard parser for adaptive-sampling QC.
Replaces the hybrid-capture QC parsers inherited from the reference pipeline
(hsmetrics, fastp, Picard), none of which apply to nanopore adaptive sampling.
The QC that does apply is what QC_ONTARGET already emits per sample:
<effective_dir>/qc/<sample>.region... | patkarlab/mm-awgs-nextflow | bin/dashboard_builder/parsers/qc.py | .py | c03be182bc8da71e | 7 | 0 |
"""
translocations.py - dashboard parser for merged, annotated structural variants.
Reads the per-sample annotated translocation table produced by
merge_translocations.py + annotate_mm_translocations.py:
<effective_dir>/translocations/<sample>.mm_annotated.tsv
The schema is read from the file's own header row; n... | patkarlab/mm-awgs-nextflow | bin/dashboard_builder/parsers/translocations.py | .py | 9a003ec2932efca3 | 7 | 0 |
"""Parse somaticseq variant TSVs.
Two flavors with overlapping schemas:
- clinical_final.tsv : curated clinical variants (smaller column set, PASS/REJECT verdict)
- filtered.tsv : full annotated set (adds ClinVar, gnomAD AF, etc.)
Both are simple tab-delimited with a header row. We use pandas for robustness... | patkarlab/mm-awgs-nextflow | bin/dashboard_builder/parsers/variants.py | .py | d9709c5cfb2299a4 | 7 | 0 |
#!/usr/bin/env python3
"""
Embedding engine for jina-embeddings-v3 with dual backend support:
1. PyTorch / Transformers (AutoModel + AutoTokenizer, trust_remote_code=True)
2. ONNX Runtime (onnx/model.onnx + onnx/model.onnx_data, task_id adapter routing, mean pooling, L2 normalization)
Supported task adapters:
- retrie... | mmahdi-sz/jina-embeddings-v3-ort | python-ref/embed.py | .py | 7e5a56725d8f6ad5 | 7 | 0 |
#!/usr/bin/env python3
"""
Comprehensive verification harness and ground-truth exporter for jina-embeddings-v3.
Verifies:
1. Structural Validity (shape 1024, float32, no NaN/Inf, unit L2 norm)
2. PyTorch vs. ONNX Numerical Parity (cosine sim > 0.9999, max diff < 1e-4)
3. Intra-Category Semantic Cohesion (pairwise simi... | mmahdi-sz/jina-embeddings-v3-ort | python-ref/verify.py | .py | e36fefdb62c92a52 | 7 | 0 |
"""Lightweight ``.env`` loader for local development (no external dependency).
Mirrors the pattern used by ``scripts/sync_notion_brain.py`` so behaviour is
identical across the repo: the gitignored project-root ``.env`` is read once
and any keys not already present in the process environment are applied.
Existing envi... | asifdotpy/forge-mind | src/forgemind/_env.py | .py | 1f7ca2d30c0f7af5 | 7 | 0 |
from __future__ import annotations
"""Small, dependency-free HTML rendering helpers."""
from typing import Any
def _esc(value: Any) -> str:
"""Minimal HTML escaping for untrusted artifact values."""
text = "" if value is None else str(value)
return (
text.replace("&", "&")
.replace("<"... | asifdotpy/forge-mind | src/forgemind/api/dashboard/helpers.py | .py | 937b8b536deed574 | 7 | 0 |
from __future__ import annotations
"""Pydantic request envelopes for the ForgeMind HTTP API."""
from typing import Any, Dict, List, Literal, Optional
from pydantic import BaseModel, ConfigDict
class EventInput(BaseModel):
"""Request envelope for ``POST /api/v1/events``.
Only ``event`` is required. The Event... | asifdotpy/forge-mind | src/forgemind/api/models.py | .py | 7d33318f451a6183 | 7 | 0 |
from __future__ import annotations
"""Pure five-tier pipeline orchestration (no HTTP logic)."""
import json
from pathlib import Path
from typing import Any, Dict, List, Optional
from forgemind._paths import FIXTURES_INPUT_DIR
from forgemind.acquisition import acquire_event
from forgemind.action_gate import ActionVali... | asifdotpy/forge-mind | src/forgemind/api/pipeline.py | .py | 77af4a914777b7f0 | 7 | 0 |
from __future__ import annotations
"""FastAPI route handlers - thin HTTP mapping over pipeline + dashboard."""
import json
import logging
from pathlib import Path
from typing import Any, Dict, Optional
from fastapi import FastAPI
from fastapi.responses import HTMLResponse, JSONResponse
from forgemind._env import loa... | asifdotpy/forge-mind | src/forgemind/api/routes.py | .py | 41e6481c9257f3f4 | 7 | 0 |
"""Gemini 3.5 (Vertex AI) adapter for bounded free-text evidence synthesis.
ADR-010 scope ONLY. This module is the single point where a model may
influence the pipeline. Its contract is deliberately narrow:
generate_observations(domain, context, *, model="gemini-3.5-flash")
-> list[str] | None
gener... | asifdotpy/forge-mind | src/forgemind/llm/adapter.py | .py | ad6b5dc1dd07bb40 | 7 | 0 |
"""M3-A judge-visible surface: pure derivation of the M3 proof block.
Presentation-only (SPEC-001 M3-A / T720). This module contains ZERO tier
logic: it reads the dict returned by :func:`forgemind.api.run_pipeline` and
projects it into the four judge-visible properties:
1. ``provenance_links`` — unbroken Event -... | asifdotpy/forge-mind | src/forgemind/m3_proof.py | .py | 517becaed1e88407 | 7 | 0 |
"""Memory tools for ForgeMind ADK agents.
Tools that agents can call to interact with long-term memory.
All tools fail gracefully when Memory Bank is not configured.
"""
from __future__ import annotations
import logging
from typing import Any, Dict, List
logger = logging.getLogger(__name__)
def remember_pattern(p... | asifdotpy/forge-mind | src/forgemind/memory/memory_tools.py | .py | a8b92deee20773b3 | 7 | 0 |
"""GitHub API client wrapper for ForgeMind ADK tools.
Provides a thin, typed wrapper around the GitHub REST API using the ``requests``
library (already a project dependency). Handles authentication, rate-limit
detection, pagination, and error normalization so individual tools stay
focused on their domain logic.
Envir... | asifdotpy/forge-mind | src/forgemind/tools/github_client.py | .py | f4c1e1637658532d | 7 | 0 |
"""Tool registry for ForgeMind ADK agents.
Provides a registry pattern for mapping tool names to tool objects
so agents can be configured with tools=get_default_registry().tools
"""
from __future__ import annotations
import logging
from typing import Any, Callable, Dict, List
logger = logging.getLogger(__name__)
... | asifdotpy/forge-mind | src/forgemind/tools/tool_registry.py | .py | 44fcf6226a2da360 | 7 | 0 |
"""Contract tests: validate the canonical Phase 0 fixtures against the machine-readable JSON Schemas.
These implement SPEC-001 Required Contract Tests (see specs/001-hierarchical-runtime-dag/spec.md):
valid event creation; provenance preservation; evidence-shard validation; escalation generation;
uncertainty preservat... | asifdotpy/forge-mind | tests/contract/test_contracts.py | .py | ffa75ba4f3032ea3 | 7.5 | 0 |
"""M3-B integration tests (T733) — bounded Gemini + ADK 2 workflow.
These tests exercise the M3-B additions WITHOUT asserting any model text:
they verify the fail-closed deterministic fallback, the ADK runtime import
surface, parity between the deterministic and ADK paths, the human-approval
pause/resume gate, and the... | asifdotpy/forge-mind | tests/contract/test_m3b_adk.py | .py | 756feffc82c6af83 | 7.5 | 0 |
"""Demo: run a tiny RLMTask through a real dspy.RLM on a Claude Pro/Max SUBSCRIPTION.
`ClaudeAgentLM` now ships in the kit — `from rlm_harness import ClaudeAgentLM`. The adapter's
setup, politeness policy, and trade-offs live in its module docstring
(`rlm_harness/claude_agent_lm.py`); this file is just the runnable de... | qazbnm456/rlm-harness | examples/claude_agent_lm.py | .py | db89e1fa16e44699 | 7.35 | 4 |
"""Example: a container-backed ISOLATED runner for ``make_command_tool``.
rlm-harness ships NO command executor on purpose — the runner's isolation IS the
security boundary (see ``rlm_harness/tools/command.py``). This example shows the
reference pattern: run each model-chosen command inside a disposable, network-off
D... | qazbnm456/rlm-harness | examples/command_runner.py | .py | b9c3f447e34eace1 | 7.35 | 4 |
"""Example: IN-PROCESS harness delegation — no subprocess, no HTTP.
``make_harness_tool`` + ``harness_from_endpoint`` (``tools/harness.py``) are transport-agnostic by
design: the kit ships no ``call_endpoint`` and never will (see their docstrings — "the kit ships
NONE and names NONE"). The two usual transports are a s... | qazbnm456/rlm-harness | examples/harness_local_run.py | .py | 09dc4ef8e1ef8efb | 7.35 | 4 |
"""Example: RLM-as-Harness — intercepted sub-LM + skills tools + traced run.
Wires every Phase A/B/C piece together (illustrative; needs real model creds and
a sandbox, so it is NOT imported by the test suite):
- a local/base model wrapped via intercept_sub_lm (validate + post-process),
- a Skills directory whose cat... | qazbnm456/rlm-harness | examples/harness_run.py | .py | ac16dd0e94b649e0 | 7.35 | 4 |
"""Worked example — serve a harness over the make_harness_tool delegation contract (the SERVER side).
Run it directly:
echo "some long context" | python examples/harness_serve.py [workdir_base]
It prints ONE `HarnessPointer` JSON line on stdout. In production the harness's OWN venv runs this as a
module (`python... | qazbnm456/rlm-harness | examples/harness_serve.py | .py | 86d7e9f9295dce76 | 7.35 | 4 |
"""Minimal real end-to-end RLM run — verifies the forward() path with a live model.
Self-contained: configures from RLM_* env, runs one tiny task through a real
dspy.RLM (real Deno sandbox + real model), records the trajectory, and prints
both the validated result and a trajectory summary so we can confirm the live
sh... | qazbnm456/rlm-harness | examples/mini_run.py | .py | 8c915d55e7d05162 | 7.35 | 4 |
"""Shims for ``dspy``'s ``RLM`` / interpreter API, resolved by introspection in ONE place.
PRIVATE (``_``-prefixed): not part of the public surface, may change without notice.
**Why this module exists, and why it survives the 3.3.0 floor.** rlm-harness declares only a
FLOOR on dspy and consumers pin the KIT, so a con... | qazbnm456/rlm-harness | rlm_harness/_dspy_compat.py | .py | 83cbc7a108cc74f9 | 7.35 | 4 |
"""The validation + retry engine shared by every RLM task.
This replaces the hand-rolled ``while execute_count < MAX_RETRY`` loops that were
copy-pasted across the original CVE app. It is deliberately free of any ``dspy``
import: it operates on a ``runner`` coroutine that returns a prediction-like
object (anything wit... | qazbnm456/rlm-harness | rlm_harness/_retry.py | .py | 8994a4db2841756e | 7.35 | 4 |
"""REPL-safety rules for a tool — its NAME and its SIGNATURE, one derivation each.
The module is ``_``-prefixed, but **three of its functions are PUBLIC and SemVer-frozen**
since 1.1.0, re-exported from ``rlm_harness.__all__``: :func:`is_valid_tool_name`,
:func:`sanitize_tool_name` and :func:`unique_tool_names`, plus
... | qazbnm456/rlm-harness | rlm_harness/_toolname.py | .py | 038d6d64d4b02946 | 7.35 | 4 |
"""``atomic_write_text`` / ``atomic_write_stream`` — write a file such that a concurrent reader
never sees a partial write.
A same-directory temp file + ``fsync`` + ``os.replace`` — the standard "never a half-written file
visible mid-write" idiom, useful for any consumer building a resumable/checkpointed job on top of... | qazbnm456/rlm-harness | rlm_harness/atomic.py | .py | 10250c8cde115dea | 7.35 | 4 |
"""Single source of truth for RLM runtime configuration.
Everything the scaffold needs to stand up a Recursive Language Model — model
names, credentials, the sandbox interpreter, budget caps, retry policy — lives
here and is driven by environment variables. No other module reads ``os.environ``.
This module intentiona... | qazbnm456/rlm-harness | rlm_harness/config.py | .py | 5ae843667b7ddcf9 | 7.35 | 4 |
"""Phase C (part 2) — export recorded runs as Agentic-RL / SFT datasets.
The JSONL trace is the source of truth. This module turns it into training-ready
records, in three shapes:
- ``export_sft_turns`` — per-root-TURN SFT samples (``input = full history`` seeded with the
run's initial state, ``output = that turn``... | qazbnm456/rlm-harness | rlm_harness/dataset.py | .py | b2fbda642fb13f47 | 7.35 | 4 |
"""``python -m rlm_harness.harness_serve <pkg.module:run> [workdir_base]`` — the zero-file way to serve a
harness over the delegation contract (the runnable front-end of :func:`rlm_harness.serve_harness`).
Resolves the harness's ``run`` callable from ``<module:attr>`` and, if the same module exposes a
``to_pointer`` (... | qazbnm456/rlm-harness | rlm_harness/harness_serve.py | .py | e014b0b7ea06c55a | 7.35 | 4 |
"""``run_in_subprocess`` — a safe, isolated-subprocess primitive.
A small PRIMITIVE only: "safely run one picklable callable in an isolated OS process, get its
result or a clear error back, bounded by a timeout." Queue/scheduling logic (how a web server
actually schedules many of these — Celery, RQ, a plain thread/pro... | qazbnm456/rlm-harness | rlm_harness/isolation.py | .py | 54c32b27e53062c0 | 7.35 | 4 |
"""Trace utilization metrics — how a run's activity was distributed across the root LM's own
turns, tool calls, and sub-LM escalations. A sibling to ``rubric.py``'s "derive facts from a trace"
shape, but structurally different: ``rubric.criteria_facts`` slices a CALLER-SUPPLIED facts dict
against a caller-supplied lens... | qazbnm456/rlm-harness | rlm_harness/metrics.py | .py | 9984718a20ff9532 | 7.35 | 4 |
"""GEPA optimization harness — PHASE 1 SKELETON.
The whole point of choosing DSPy for RLM is that tasks can be *compiled*
(prompt + few-shot demos optimised against a metric) rather than hand-tuned.
This module wires that interface and ships ready-to-use metric templates.
What is implemented now (Phase 1):
- Metric t... | qazbnm456/rlm-harness | rlm_harness/optimize.py | .py | 9ce0aae5931c5a86 | 7.35 | 4 |
"""Phase C (part 1) — reconstruct and replay a recorded run.
Replay reads the JSONL trace and rebuilds an ordered timeline. For deterministic
replay it serves *recorded* tool outputs rather than re-executing tools (which
may be non-deterministic or have side effects). This makes a past run inspectable
and step-through... | qazbnm456/rlm-harness | rlm_harness/replay.py | .py | 18edec98efaa01a3 | 7.35 | 4 |
"""Reward-free rubric primitives — the shared substrate for decomposing "did this run succeed?" into
observable CRITERIA carried as LABELS.
``category`` is an OPAQUE, caller-defined label: rlm-harness never interprets it, hardcodes no taxonomy, and
carries no domain vocabulary. A consumer defines its own category set,... | qazbnm456/rlm-harness | rlm_harness/rubric.py | .py | a140d1bf28486ca8 | 7.35 | 4 |
"""One-time runtime initialization: wire dspy + (optional) observability.
Replaces the original app's scattered ``agent.py`` global setup. Call
:func:`configure` once at process start; tasks then read the shared config and
sub-LM via :func:`get_config` / :func:`get_sub_lm`.
"""
from __future__ import annotations
imp... | qazbnm456/rlm-harness | rlm_harness/runtime.py | .py | d818eda09f689812 | 7.35 | 4 |
"""Serve an rlm-harness harness over the delegation contract — the SERVER-side mirror of
``make_harness_tool`` (``tools/harness.py``).
``make_harness_tool`` is the CLIENT: a parent RLM wraps a downstream harness as a tool and reaches it
through an injected ``call_endpoint`` (a subprocess command, an HTTP URL, …). This... | qazbnm456/rlm-harness | rlm_harness/serving.py | .py | 66f6deebc36d0dbe | 7.35 | 4 |
"""Phase A — expose a directory of Skills to the RLM as tools.
A "Skill" here follows the common convention of a folder containing a ``SKILL.md``
(with optional YAML-ish frontmatter for ``name``/``description``), or a flat
``<name>.md`` file. The main LM decides, inside the REPL, which skill to read —
keeping control ... | qazbnm456/rlm-harness | rlm_harness/skills.py | .py | e8791d4e9615d4ee | 7.35 | 4 |
"""Phase A — ``intercept_sub_lm``: the one hook to intercept the RLM's sub-LM.
``dspy.RLM`` exposes no hook to intercept a sub-LLM response before it returns to
the main model — and its built-in ``llm_query`` / ``llm_query_batched`` tools just
call ``self.sub_lm(prompt)``. So the ONLY interception point is the sub_lm ... | qazbnm456/rlm-harness | rlm_harness/sub_lm.py | .py | 49d8c09b822bbd24 | 7.35 | 4 |
"""The ``RLMTask`` base class — the one abstraction this scaffold exists for.
A task is declared by subclassing ``RLMTask`` and filling four fields:
class Summarize(RLMTask):
signature = "document: str -> article: Article"
output_field = "article"
output_model = Article # a... | qazbnm456/rlm-harness | rlm_harness/task.py | .py | 00700c6a280384af | 7.35 | 4 |
"""Test support for driving the RLM forward path OFFLINE — no live model, no Deno, no network.
``dspy.RLM`` normally runs the model's Python inside a sandboxed interpreter (pyodide/deno). That makes
the *forward* path (planner turn -> tool call -> SUBMIT -> validated result) expensive to test: it needs
a paid model an... | qazbnm456/rlm-harness | rlm_harness/testing.py | .py | d79275ab2d6dd47b | 7.85 | 4 |
"""``run_isolated`` — bridge a coroutine into a sync call site that may already own a running
event loop (mirrors the bridging problem ``mcp.py`` solves for its own, different reason).
A consumer building their OWN transport for :func:`rlm_harness.tools.harness_from_endpoint` (e.g.
an in-process ``call_endpoint`` that... | qazbnm456/rlm-harness | rlm_harness/tools/_async.py | .py | ebbcda43ff10d129 | 7.35 | 4 |
"""``make_extract_archive_tool`` — safe zip/tar extraction into a bounded local directory.
Python's ``zipfile.extractall()``/``tarfile.extractall()`` are not safe by default: a malicious
archive entry can carry an absolute path, a ``..``-traversal path, or (tar) a symlink/hardlink
pointing outside the extraction targe... | qazbnm456/rlm-harness | rlm_harness/tools/archive.py | .py | 20faee6420fd19af | 7.35 | 4 |
"""Reusable ``run_command`` tooling — execute a local command through a
consumer-supplied, ISOLATED runner (mirrors ``fetch.py`` / ``search.py``).
An agent built on the RLM often needs to run a local command (a build, a test, a
git op) the way a coding agent does. The reusable half is the same as every other
tool here... | qazbnm456/rlm-harness | rlm_harness/tools/command.py | .py | 04c445ab1b0813b7 | 7.35 | 4 |
"""``list_candidate_paths`` — a safe, good-default way to compute ``make_read_file_tool``/
``make_grep_files_tool``'s ``candidate_paths``.
``fs.py`` deliberately leaves ``candidate_paths`` REQUIRED, consumer-supplied — no default
directory walk, no built-in ``.gitignore`` handling (the base/wrap split: the kit owns th... | qazbnm456/rlm-harness | rlm_harness/tools/discover.py | .py | b5b30f4f19fe5122 | 7.35 | 4 |
"""``make_write_file_tool`` / ``make_edit_file_tool`` — the write side of the filesystem tools,
sitting alongside ``fs.py``'s read side (``make_read_file_tool`` / ``make_grep_files_tool``).
Kept in a SEPARATE module from ``fs.py`` (which is already the largest single file in `tools/` —
355 lines, ~1.6× the next-larges... | qazbnm456/rlm-harness | rlm_harness/tools/edit.py | .py | bd2bff41dbb2c7cb | 7.35 | 4 |
"""SSRF-aware fetch tooling for RLM tasks.
Tasks routinely need to pull remote content (web pages, docs, feeds, threat
intel). Handing an LLM-driven REPL an unrestricted fetcher is an SSRF liability:
the model can be steered — by the very untrusted content it is analysing — into
requesting internal services or cloud m... | qazbnm456/rlm-harness | rlm_harness/tools/fetch.py | .py | 39ff4fb3a4299e1a | 7.35 | 4 |
"""``make_git_clone_tool`` — safe git clone with fallback auth, base/wrap, same shape as
``make_fetch_tool``/``make_command_tool`` (not a new pattern).
A task that wants to clone a repository to analyze it has no safe way to do so without this: (1)
SSRF-shaped URL abuse (a ``file://`` URL, an internal git server); (2)... | qazbnm456/rlm-harness | rlm_harness/tools/git_clone.py | .py | 667a862adcb682b8 | 7.35 | 4 |
"""Provider-agnostic ``make_harness_tool`` — delegate a sub-task to ANOTHER rlm-harness harness, wrapped
as a tool (the promoted "wrap a downstream harness as a tool" shape; mirrors ``model.py``).
A *harness* is a full RLM in its own right: it takes a long-text input, runs its own Root LM in a REPL
loop over that text... | qazbnm456/rlm-harness | rlm_harness/tools/harness.py | .py | 8259d2e2cb88cd48 | 7.35 | 4 |
"""Provider-agnostic ``make_model_tool`` — the generic "model-backed tool + validate"
core (mirrors ``fetch.py`` / ``search.py``).
A model-as-tool — a SECONDARY model the RLM root calls as a tool to PRODUCE something
(YAML, code, SQL, …) which is then deterministically validated — is a recurring shape.
The reusable me... | qazbnm456/rlm-harness | rlm_harness/tools/model.py | .py | ef613ec55a4aa649 | 7.35 | 4 |
"""Provider-agnostic ``web_search`` building blocks (mirrors ``fetch.py``).
A search tool needs two halves: the PROVIDER (an HTTP call to DuckDuckGo / Tavily /
TinyFish / … plus its API key) and the generic GUARD/NORMALISE step. rlm-harness owns only
the generic half — it picks NO provider. The consuming project suppl... | qazbnm456/rlm-harness | rlm_harness/tools/search.py | .py | 7561169d54d65a76 | 7.35 | 4 |
"""Schema-validation tool factories.
Two shapes, both consumer-facing base primitives:
- ``make_schema_validator(model)`` — a plain callable the RLM invokes inside the REPL to
check its draft JSON against a pydantic schema before emitting a final answer (returns a
human message). The generalised form of the origi... | qazbnm456/rlm-harness | rlm_harness/tools/validation.py | .py | 138570a16d600ba0 | 7.35 | 4 |
"""`run_isolated` — the async-bridge primitive for a consumer's own harness-delegation transport.
All offline, no dspy: `run_isolated` is pure `threading`/`asyncio`, dspy-free by construction."""
import asyncio
import pytest
from rlm_harness.tools import run_isolated
from rlm_harness.trace import TraceRecorder, curr... | qazbnm456/rlm-harness | tests/test_async.py | .py | 9b32b57722d78583 | 7.85 | 4 |
"""atomic_write_text / atomic_write_stream — write-without-partial-file. All offline, dspy-free."""
from __future__ import annotations
import os
import stat
import pytest
from rlm_harness import atomic_write_stream, atomic_write_text
from rlm_harness.atomic import _ExtractionBudgetExceeded
def _read(path):
wit... | qazbnm456/rlm-harness | tests/test_atomic.py | .py | fd49fb56153511d8 | 7.85 | 4 |
"""ClaudeAgentLM tests — the optional Claude-subscription adapter (`rlm-harness[subscription]`).
The heavy `claude-agent-sdk` is NOT a test dependency: the pure helpers run without it, the
lazy export is asserted without it, and construction is exercised against a FAKE
`claude_agent_sdk` injected into `sys.modules` — ... | qazbnm456/rlm-harness | tests/test_claude_agent_lm.py | .py | ae0575572c4f88af | 7.85 | 4 |
"""Verify the CHANGELOG compare links match the declared version.
Cutting a release moves entries into a new section, and the link definitions
live ~1700 lines away at the bottom of the file — so the cut looks complete on
screen while the links still point at the previous release. That has now
happened twice: PR #100 ... | andrei-shtanakov/spec-runner | scripts/check_changelog_links.py | .py | c15b87cd2f3a0d2f | 7 | 0 |
"""Compliance audit-trail writer (LABS-40).
Structured JSON-Lines appender for regulated projects that need a durable
record of every state transition the executor made, independent of the
ordinary structlog output (which is console-oriented and rotates).
Opt-in: disabled by default. Enable by setting `audit_log_path... | andrei-shtanakov/spec-runner | src/spec_runner/audit_log.py | .py | 1559e091531e0754 | 7.5 | 0 |
"""The status flip a blocked task leaves behind, committed — #192 (F-8).
`post_done_hook` writes `🔍 REVIEW` into `tasks.md` when review *starts*, so a
run killed mid-review is resumable rather than prematurely DONE (#66). That
write is uncommitted by design: the commit that would carry it comes later. But
when a pre-... | andrei-shtanakov/spec-runner | src/spec_runner/bookkeeping.py | .py | 715c03ebf9ee1c3d | 7.5 | 0 |
"""The pre-call budget guard (#213, second half).
**A guard, not a cap.** It answers one question, immediately before a paid
call: *is there anything left to spend?* It cannot stop the call that crosses
the line, because a call's cost is known only after it returns. The only true
hard cap is a backend-enforced per-cal... | andrei-shtanakov/spec-runner | src/spec_runner/budget.py | .py | f9090a092abb792a | 7.5 | 0 |
"""`spec-runner budget authorize` — an operator raising a ceiling (#230 part 2).
Refunds and a separate infrastructure budget were rejected at design time: both
stop the sentence *"the number bounds the money"* from being true. A refund
turns the cap into a progress bound, and a deterministic instrument failure
then l... | andrei-shtanakov/spec-runner | src/spec_runner/budget_cmd.py | .py | 582d01011d5703b6 | 7.5 | 0 |
"""``change`` subcommands: new, list, archive (M2, change-as-folder).
A change lives at ``spec/changes/<id>/`` and is a self-rooted spec dir — the
rest of the toolchain scopes to it via ``config.change_id`` (CLI ``--change``).
Archiving here only moves the folder to ``spec/changes/archive/`` with a date
prefix; mergin... | andrei-shtanakov/spec-runner | src/spec_runner/change_commands.py | .py | eb6643067b2ea74d | 7.5 | 0 |
"""File claims — the byte-lock behind a confirmed RED (#141 slice 2).
A claim says: *this file, at these bytes, is frozen, because a confirmed RED
depends on it.* The pilot's first version checked only the file of the current
selector, so neighbouring tests were protected by a sentence in the agent's
prompt rather tha... | andrei-shtanakov/spec-runner | src/spec_runner/claims.py | .py | 33f99a4af384851b | 7.5 | 0 |
"""Error classification for CLI agent stderr (v2.3.0).
Adds short, human-readable reasons to failures (previously surfaced as
"Unknown error"). Pattern library + last-N-lines stderr fallback.
"""
from __future__ import annotations
import re
from dataclasses import dataclass
STDERR_TAIL_LINES = 5
@dataclass(frozen... | andrei-shtanakov/spec-runner | src/spec_runner/errors.py | .py | 1ae1754a9acd0091 | 7.5 | 0 |
"""Event bus for streaming task execution events to TUI and other subscribers."""
import asyncio
import collections
import contextlib
import threading
import time
from dataclasses import dataclass, field
@dataclass
class TaskEvent:
"""A single event from task execution."""
task_id: str
event_type: str ... | andrei-shtanakov/spec-runner | src/spec_runner/events.py | .py | fc96ac3df5e685ef | 7.5 | 0 |
"""Backward-compatible re-exports.
All public API is available from this module for existing imports.
Implementation moved to execution.py, cli.py.
"""
from .logging import get_logger
logger = get_logger("executor")
# Global shutdown flag — kept here because state.py imports it from .executor
_shutdown_requested = ... | andrei-shtanakov/spec-runner | src/spec_runner/executor.py | .py | 5d883ab7d6cf9a73 | 7.5 | 0 |
"""Harness-mutation tripwire (#64).
The verification harness (test/lint configuration, dependency manifests, CI
workflows) lives inside the agent's write scope: an agent that sees a failing
gate can create a bridge — or neuter the oracle outright — and the run
reports success. Observed in the field: a pytest bridge (`... | andrei-shtanakov/spec-runner | src/spec_runner/harness.py | .py | 384a34fa3ec417ff | 7.5 | 0 |
"""
spec-runner init — install Claude Code skills to project.
Usage:
spec-runner-init # Install to .claude/skills in current directory
spec-runner-init /path/to/project
"""
from __future__ import annotations
import argparse
import shutil
import sys
from pathlib import Path
def get_skills_sourc... | andrei-shtanakov/spec-runner | src/spec_runner/init_cmd.py | .py | 788baaabe45f8f8d | 7.5 | 0 |
"""The TDD lifecycle as a recorded state machine — #141 slice 4a.
Slices 1–3 built the parts: a verified red, a byte-lock on what it depends on,
typed operator remedies. Where a task *was* still lived in inference — read the
checkpoints, read the claims, guess. This makes it a fact.
```
READY → RED_AUTHORING → RED_VE... | andrei-shtanakov/spec-runner | src/spec_runner/lifecycle.py | .py | 5d922cbb69647316 | 7.5 | 0 |
"""Back-compat shim over spec_runner.obs.
The canonical entrypoint is now `spec_runner.obs.init_logging`. This module
remains for existing callers that import `setup_logging`, `get_logger`, or
`redact_sensitive`.
"""
from __future__ import annotations
import re
from pathlib import Path
import structlog
from spec_r... | andrei-shtanakov/spec-runner | src/spec_runner/logging.py | .py | 9cae769429c1fd70 | 7.5 | 0 |
"""MCP server for spec-runner -- exposes status, tasks, costs, logs, and execution tools.
Security: the stdio transport inherits the trust boundary of the process that
launched it (typically a developer's terminal or Claude Code). There is no
built-in authentication. Write tools (`run_task`, `stop`) spawn subprocesses... | andrei-shtanakov/spec-runner | src/spec_runner/mcp_server.py | .py | 3c015c90f35a41ae | 7.5 | 0 |
"""Notifications for spec-runner.
Sends notifications via Telegram Bot API and/or generic webhook
on run_complete, task_failed, state_degraded and pr_opened events.
Best-effort — errors are logged, never raised.
Notifications are ONLY sent when explicitly configured in the
project config file (spec-runner.config.yaml... | andrei-shtanakov/spec-runner | src/spec_runner/notifications.py | .py | 7e101fe7e1e27811 | 7.5 | 0 |
"""Orchestra observability emitter — reference implementation.
Source of truth for `obs.py` (vendored into other Python projects).
Produces OpenTelemetry Logs Data Model JSONL, one file per PID.
Contract: see maestro/contracts/observability/log-schema.json
"""
from __future__ import annotations
import logging as _s... | andrei-shtanakov/spec-runner | src/spec_runner/obs.py | .py | 0263a47cf5e2088d | 7.5 | 0 |
"""Typed outcome for a task phase — slice 0 of the lifecycle contract.
Until now a stage said only where it was (`StageReporter`) and, if it died,
where it died (`attempts.error_stage`). That is the whole vocabulary: a stage
either fell over or it did not. One phase already grew a real one under
pressure — `review`, i... | andrei-shtanakov/spec-runner | src/spec_runner/phases.py | .py | 5e9e2485c70c226f | 7.5 | 0 |
"""Plugin discovery and loading for spec-runner.
Scans a plugins directory for subdirectories containing plugin.yaml
manifests, parses them into PluginInfo/PluginHook dataclasses.
Executes plugin hooks as subprocesses with env vars and run_on filtering.
"""
from __future__ import annotations
import os
import subproc... | andrei-shtanakov/spec-runner | src/spec_runner/plugins.py | .py | 56bddf614386b394 | 7.5 | 0 |
"""Read-only preflight: what is missing, and what of that blocks a run (#142a).
There is no zero stage. On a greenfield repo "what do I need before tasks can
run" was answered one task failure at a time, and a gate that is green on an
empty project answers nothing at all: an empty suite exits 0, and so does a
linter w... | andrei-shtanakov/spec-runner | src/spec_runner/preflight.py | .py | 2540617f7655956f | 7.5 | 0 |
"""spec-runner config — apply CLI profile presets to spec-runner.config.yaml."""
from __future__ import annotations
import argparse
import shutil
import sys
from dataclasses import dataclass
from importlib.resources import files
from pathlib import Path
from typing import TYPE_CHECKING
import yaml
if TYPE_CHECKING:... | andrei-shtanakov/spec-runner | src/spec_runner/preset_cmd.py | .py | 83331eab97456df5 | 7.5 | 0 |
"""One writer for the prompts of the paid calls (#282, and the review half after it).
Three stages of a task cost money — RED authoring, the implementation pass, and
review — and until #282 only the implementation pass left a record of what it
had asked. The RED prompt decides which file is frozen for the rest of the
... | andrei-shtanakov/spec-runner | src/spec_runner/prompts_log.py | .py | 63656a5dd9c123ca | 7.5 | 0 |
"""Traceability matrix report.
Maps requirements -> design decisions -> tasks -> execution state
to produce a full pipeline visibility report.
"""
from __future__ import annotations
import json
import re
from dataclasses import dataclass, field
from .config import ExecutorConfig
from .state import ExecutorState
fro... | andrei-shtanakov/spec-runner | src/spec_runner/report.py | .py | 0bd9552187538dd0 | 7.5 | 0 |
"""Structured, tolerant parsing of requirements documents (M1).
Turns a ``requirements.md`` into id-keyed :class:`Requirement` blocks so a
requirement becomes a diffable/mergeable unit — the foundation for delta specs
and archive merge (M3).
Requirements in the wild use heterogeneous sub-structure (gherkin acceptance... | andrei-shtanakov/spec-runner | src/spec_runner/requirements.py | .py | 95efa57fc32283e5 | 7.5 | 0 |
"""`spec` subcommands: status, approve, reject, adopt, check."""
from __future__ import annotations
import argparse
import subprocess
from collections.abc import Callable
from datetime import UTC, datetime
from .config import ExecutorConfig, ExecutorLock
from .logging import get_logger
from .spec import (
SpecMe... | andrei-shtanakov/spec-runner | src/spec_runner/spec_commands.py | .py | 88cfbaaf11a90e67 | 7.5 | 0 |
"""Deterministic merge of a delta spec into the source-of-truth requirements (M3).
Identity is the REQ/NFR id, so matching is exact (no whitespace-tolerant header
heuristics as in OpenSpec): ADDED requires a new id, MODIFIED replaces the
whole existing block, REMOVED deletes it (Reason + Migration are mandatory),
RENA... | andrei-shtanakov/spec-runner | src/spec_runner/spec_merge.py | .py | 2040ec5a7d67546f | 7.5 | 0 |
"""Per-task sub-stage tracking and mirroring (v2.3.0).
One StageReporter per task. Threaded explicitly through execution; safe
with `max_concurrent > 1` because each task gets its own reporter and
there are no thread-locals.
"""
from __future__ import annotations
from collections.abc import Callable
from typing impo... | andrei-shtanakov/spec-runner | src/spec_runner/stages.py | .py | 7628c800315c6ca9 | 7.5 | 0 |
"""Post-merge sync command (#73): close the run → PR → merge → next-run loop.
After the human merges the integration PR, the operator used to hand-run
`git pull --ff-only`, delete merged run/task branches locally and on the
remote, and sanity-check executor state. `spec-runner sync` does exactly
that, reporting each s... | andrei-shtanakov/spec-runner | src/spec_runner/sync_cmd.py | .py | 921e407cab199115 | 7.5 | 0 |
# jinxurn.py
"""
Main module for JinxUrn application.
"""
import argparse
import logging
import sys
from typing import Optional
class JinxUrn:
"""Main class for JinxUrn functionality."""
def __init__(self, verbose: bool = False):
"""Initialize with verbosity setting."""
self.verbose = ver... | annakowqmvd/JinxUrn | jinxurn.py | .py | decac1defff464ec | 7 | 0 |
# test_jinxurn.py
"""
Tests for JinxUrn module.
"""
import unittest
from jinxurn import JinxUrn
class TestJinxUrn(unittest.TestCase):
"""Test cases for JinxUrn class."""
def test_initialization(self):
"""Test class initialization."""
instance = JinxUrn()
self.assertIsInstance(inst... | annakowqmvd/JinxUrn | test_jinxurn.py | .py | 34c8cd92fe4891be | 7.5 | 0 |
import os
import numpy as np
import pandas as pd
# Define fixed crime categories for consistent encoding across train/val/test
CRIME_CATEGORIES = [
'routine_transaction',
'suspicious_cash_withdrawal',
'unusual_online_activity',
'high_value_transfer'
]
def extract_features(df):
"""
Transforms p... | malikafarah/CyberSentinel | ml-model/src/features.py | .py | fbe9727538af4fbf | 7 | 0 |
import os
import joblib
import numpy as np
import pandas as pd
from typing import List, Optional
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
# 1. Define Request / Response Schemas (API Contract)
class LocationCandidate(BaseModel):
location_id: str
latitude: float
longitude: fl... | malikafarah/CyberSentinel | ml-model/src/predictionapi.py | .py | adefb2f03f3a15eb | 7 | 0 |
import os
import numpy as np
import pandas as pd
# City-to-Coordinate Mapping for the 43 US Cities in the dataset
CITY_COORDINATES = {
'Albuquerque': (35.0844, -106.6504),
'Atlanta': (33.7490, -84.3880),
'Austin': (30.2672, -97.7431),
'Baltimore': (39.2904, -76.6122),
'Boston': (42.3601, -71.0589),... | malikafarah/CyberSentinel | ml-model/src/preprocessing.py | .py | 5c267895fd778d32 | 7 | 0 |
#!/usr/bin/env python3
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Trivial sidecar k8s charm used to validate the opcli CI build pipeline.
Deploys to a k8s model with a rock container. Goes active once pebble is ready.
"""
import ops
class K8sCharm(ops.CharmBase):
"""A minimal ... | canonical/charm-ci | examples/k8s-charm/src/charm.py | .py | 4d213ee9b3608cdd | 7.3 | 3 |
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Fixtures for k8s-charm sub-charm integration tests.
This demonstrates a monorepo layout where a sub-charm has its own
integration tests directory with its own conftest, independent of the
top-level tests/integration/.
Artifact fixtures (cha... | canonical/charm-ci | examples/k8s-charm/tests/integration/conftest.py | .py | 9a49fadc3428484a | 7.8 | 3 |
#!/usr/bin/env python3
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Trivial machine charm used to validate the opcli CI build pipeline."""
import ops
class MachineCharm(ops.CharmBase):
"""A minimal machine charm that immediately goes active."""
def __init__(self, *args: obje... | canonical/charm-ci | examples/machine-charm/src/charm.py | .py | 506e9274d905da52 | 7.3 | 3 |
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Fixtures for machine-charm sub-charm integration tests.
Artifact fixtures (charm_path, etc.) are provided automatically by the
pytest-opcli plugin — no flag plumbing needed.
"""
from collections.abc import Generator
import jubilant
import ... | canonical/charm-ci | examples/machine-charm/tests/integration/conftest.py | .py | b1742f69b9d49159 | 7.8 | 3 |
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Integration test for the machine-charm sub-charm (monorepo pattern).
Demonstrates auto-discover: false — variants are listed explicitly.
"""
import jubilant
from opcli.pytest_plugin import CharmPathList
def test_deploy(juju: jubilant.Juj... | canonical/charm-ci | examples/machine-charm/tests/integration/test_charm.py | .py | 92a69d5089a369f3 | 7.8 | 3 |
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Shared pytest fixtures and CLI options for examples integration tests.
Artifact fixtures (charm_paths, resource_images, etc.) are provided
automatically by the pytest-opcli plugin — no flag plumbing needed.
For multi-charm repos ``resource_... | canonical/charm-ci | examples/tests/integration/conftest.py | .py | 17b39eb80a6dcf1c | 7.8 | 3 |
# Copyright 2024 Canonical Ltd.
# See LICENSE file for licensing details.
"""Integration test: machine-charm deploys and reaches active/idle."""
import os
import jubilant
from opcli.pytest_plugin import CharmPathList
def test_spread_job_forwarded_to_pytest() -> None:
"""SPREAD_JOB set by spread is forwarded i... | canonical/charm-ci | examples/tests/integration/test_machine_charm.py | .py | e51dbdd502d89376 | 7.8 | 3 |
# Copyright 2026 Canonical Ltd.
# See LICENSE file for licensing details.
"""Top-level Typer application — registers all command groups."""
import logging
import sys
from collections.abc import Sequence
from typing import Any
try:
import typer
from typer.core import TyperGroup
except ImportError:
print(
... | canonical/charm-ci | src/opcli/app.py | .py | 2c39fcccd44c58e8 | 7.3 | 3 |
# Copyright 2026 Canonical Ltd.
# See LICENSE file for licensing details.
"""CLI commands for artifact discovery and building."""
import json
from pathlib import Path
from typing import Annotated
import typer
from opcli.core.artifacts import (
_DEFAULT_WAIT_TIMEOUT_SECONDS,
artifacts_build,
artifacts_co... | canonical/charm-ci | src/opcli/commands/artifacts.py | .py | d293179c0199794f | 7.3 | 3 |
# Copyright 2026 Canonical Ltd.
# See LICENSE file for licensing details.
"""CLI commands for test environment management."""
from pathlib import Path
import typer
from opcli.core.provision import provision_prepare, provision_registry
app = typer.Typer(
help="Manage test environments (provisioning and registry... | canonical/charm-ci | src/opcli/commands/env.py | .py | e08d77427a20f11a | 7.3 | 3 |
# Copyright 2026 Canonical Ltd.
# See LICENSE file for licensing details.
"""CLI commands for installing tool dependencies."""
import typer
from opcli.core.install import (
install_all,
install_charmcraft,
install_concierge,
install_doctor,
install_gh,
install_lxd,
install_rockcraft,
... | canonical/charm-ci | src/opcli/commands/install.py | .py | d581a97db59a7d77 | 7.3 | 3 |
# Copyright 2026 Canonical Ltd.
# See LICENSE file for licensing details.
"""CLI commands for spread-based test execution."""
import json
from pathlib import Path
from typing import Annotated
import typer
from opcli.core.spread import spread_expand, spread_init, spread_jobs, spread_run
app = typer.Typer(
help=... | canonical/charm-ci | src/opcli/commands/spread.py | .py | 6c3aefec96f559d5 | 7.3 | 3 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.