pyproject.toml

[project]
name = "backgroundagent"
version = "0.1.0"
description = "Background coding agent — runs tasks in isolated cloud environments and produces pull requests"
requires-python = ">=3.13"
dependencies = [
    "boto3==1.43.38", #https://pypi.org/project/boto3/
    # Vestigial from the parked AgentCore Identity flow (Phase 2.0a).
    # Phase 2.0b reads per-workspace Linear OAuth tokens directly from
    # Secrets Manager because AgentCore Identity's USER_FEDERATION
    # flow has an open service-side bug (see memory/project_oauth_2_0b.md).
    # Kept here so the workload-token bridge in `server.py` still
    # imports cleanly when Phase 2.0c eventually resumes the
    # AgentCore Identity path. The bridge is now wrapped in
    # try/except (ImportError, AttributeError), so removing this dep
    # would degrade gracefully — but for now we keep the dep to
    # preserve the clean code path.
    "bedrock-agentcore==1.16.0", #https://pypi.org/project/bedrock-agentcore/
    "claude-agent-sdk==0.2.110", #https://github.com/anthropics/claude-agent-sdk-python/releases/tag/v0.2.110 (bundles claude CLI 2.1.191; kept in lockstep with the npm CLI pin in the Dockerfile, #215)
    "requests==2.34.2", #https://pypi.org/project/requests/
    "fastapi==0.139.0", #https://pypi.org/project/fastapi/
    "uvicorn==0.49.0", #https://pypi.org/project/uvicorn/
    "aws-opentelemetry-distro==0.18.0", #https://pypi.org/project/aws-opentelemetry-distro/
    "mcp==1.28.1", #https://pypi.org/project/mcp/
    # CEDAR ENGINE PARITY — DO NOT BUMP IN ISOLATION.
    # cedarpy (Python, agent runtime) and @cedar-policy/cedar-wasm (TypeScript,
    # CDK Lambdas) are two language bindings over the same Cedar Rust core.
    # Even patch-version drift between the bindings can produce divergent
    # (decision, matching_rule_ids) on the same (policy, input) — a class
    # of bug invisible to per-side unit tests. The contracts/cedar-parity/
    # golden fixtures are how CI catches divergence; if you bump cedarpy
    # you MUST bump @cedar-policy/cedar-wasm to a tested-compatible version
    # in cdk/package.json AND refresh the parity fixtures, in the same
    # commit. See docs/design/CEDAR_HITL_GATES.md §15.6 (decision #23) and
    # the parity-contract banner in mise.toml.
    # EXACT pin (no ^/~). The binding version (4.8.4) is the cedarpy package
    # release, NOT the Cedar Rust core version — it differs from the TypeScript
    # binding @cedar-policy/cedar-wasm (pinned at 4.8.2 in cdk/package.json).
    # Matching binding version *strings* across languages is neither necessary
    # nor sufficient for behavioral parity; parity is established empirically by
    # the contracts/cedar-parity/ golden fixtures in CI, which assert identical
    # (decision, matching_rule_ids) for both bindings on the same (policy, input).
    "cedarpy==4.8.4", #https://github.com/k9securityio/cedar-py
    # Workflow-driven tasks (#248): the step runner loads YAML workflow files
    # and validates them against agent/workflows/schema/workflow.schema.json.
    # Both were previously only transitively present; declared directly so the
    # workflow loader does not depend on another package's transitive pin.
    "pyyaml==6.0.3", #https://pypi.org/project/PyYAML/
    "jsonschema==4.26.0", #https://pypi.org/project/jsonschema/
]

[tool.uv]
constraint-dependencies = [
    "pyjwt>=2.13.0",  # PYSEC-2026-175/177/178/179 — transitive via mcp; remove when mcp bumps floor (#267)
    "pydantic-settings>=2.14.2",  # GHSA-4xgf-cpjx-pc3j — transitive via mcp; remove when mcp bumps floor
]

# Dead-code detection (#282, cairn MVG gate #6). ruff "F" catches unused
# imports/locals; vulture catches unused module-level functions/classes/methods
# that ruff cannot see. min_confidence 80 keeps the signal high; intentional
# keeps (e.g. SDK-mandated hook params) live in .vulture_allowlist.py.
[tool.vulture]
paths = ["src", ".vulture_allowlist.py"]
min_confidence = 80

[tool.bandit]
exclude_dirs = ["tests", ".venv"]
skips = [
    "B101",  # assert_used — tests use assert, ruff S101 handles per-file
    "B602",  # subprocess shell=True — needed for shell tool execution
    "B603",  # subprocess calls — needed for shell tool execution
    "B607",  # partial executable paths — sh -c pattern is safe
    "B701",  # jinja2_autoescape_false — templates are LLM prompts, not web HTML
]


[dependency-groups]
dev = [
    "ruff",
    "ty",
    "pytest",
    "pygments==2.20.0",
    "pytest-cov==7.1.0",
    "pytest-timeout==2.4.0",  # per-test wall-clock cap: a single hung test (network/subprocess/Bedrock without its own timeout) must fail LOUDLY with a traceback, not silently burn the whole build-verify budget (ABCA-684/686: one hang stalled the baseline build past its 3600s ceiling)
    "vulture==2.16",  # dead-code detection (#282): unused functions/classes ruff F can't see
]

[tool.ruff]
target-version = "py313"
line-length = 100
# The vulture allowlist is bare-name expressions by design (that is how vulture
# "uses" a name to suppress it). ruff would flag every line F821/B018, so it is
# not Python source ruff should lint. See .vulture_allowlist.py (#282).
extend-exclude = [".vulture_allowlist.py"]

[tool.ruff.lint]
select = [
    "E",     # pycodestyle errors
    "W",     # pycodestyle warnings
    "F",     # pyflakes
    "I",     # isort
    "B",     # flake8-bugbear
    "S",     # flake8-bandit (security)
    "UP",    # pyupgrade
    "SIM",   # flake8-simplify
    "TCH",   # flake8-type-checking
    "RUF",   # ruff-specific rules
    # AI007 guard (#258): magic-value comparisons should use named constants.
    # Values shared across Python/TypeScript belong in contracts/constants.json
    # (see contracts/constants.md).
    "PLR2004",  # pylint magic-value-comparison
]
ignore = [
    "S603",  # subprocess call — allowed for agent CLI operations
    "S607",  # partial executable path — agent relies on PATH for git, gh, mise
]

[tool.ruff.lint.per-file-ignores]
"src/entrypoint.py" = ["E402"]  # re-export shim: importlib.reload() call must precede re-export from-imports
"src/prompts/*.py" = ["E501"]  # long prompt strings
"tests/**" = ["S101", "S106", "S108", "E402", "PLR2004"]  # assert; test tokens; /tmp paths; importorskip; literal assertions

[tool.pytest.ini_options]
testpaths = ["tests"]
pythonpath = ["src"]
# pytest-timeout: hard wall-clock cap PER TEST. A unit test that blocks (an
# unmocked network/subprocess/Bedrock call with no timeout of its own) otherwise
# hangs the whole `mise run build` until the 3600s build-verify ceiling kills it,
# turning one flaky test into an un-diagnosable 60-min stall (ABCA-684/686). With
# this, the offending test fails with a dumped stack in 120s and the suite moves on.
# method="signal" (SIGALRM) actually ABORTS the hung test — pytest runs
# single-threaded in the main thread here, so the alarm interrupts even a blocked
# C-level/socket call. method="thread" only PRINTS the stack and cannot interrupt a
# syscall-blocked test (proven live on ABCA-688: the thread method dumped the stack
# but the build kept hanging ~55 min until the 3600s ceiling). signal is the real fix.
timeout = 120
timeout_method = "signal"

[tool.coverage.run]
branch = true
source = ["src"]
relative_files = true

[tool.coverage.report]
fail_under = 72
precision = 2
show_missing = true
skip_covered = true

[tool.ty.rules]
# asyncio.get_event_loop_policy() is deprecated in 3.14+; still valid on 3.13.
deprecated = "ignore"

[tool.ty.environment]
python-version = "3.13"
extra-paths = ["src"]

[tool.ty.src]
# The vulture allowlist is bare-name expressions by design, not type-checkable
# Python — ty flags each line unresolved-reference. Excluded here for the same
# reason ruff excludes it via extend-exclude (#282).
exclude = [".vulture_allowlist.py"]