Skip to content

CLAUDE.md

Latest commit

 

History

History
66 lines (47 loc) · 4.92 KB

CLAUDE.md

File metadata and controls

66 lines (47 loc) · 4.92 KB

Claude Code Instructions

Before Committing

Always run these commands before committing changes:

# Run type checking
cd src && uv run pyright .

# Run tests
cd src && uv run pytest -q

# Run all pre-commit hooks (linting, formatting, terraform docs, etc.)
uv run --directory src --extra dev pre-commit run --all-files

Running Tests

cd src && uv run pytest -q

Architecture Overview

Terraform module that deploys AWS Lambda functions for just-in-time AWS SSO access management via Slack. Two access models exist: account access (temporary permission set assignment on AWS accounts) and group access (temporary SSO group membership).

Terraform Layer

  • vars.tf — Module inputs. Two separate config variables: config (account access statements) and group_config (group access statements), both type = any.
  • s3.tf — Serializes both to a single S3 object as {"statements": var.config, "group_statements": var.group_config}.
  • locals.tf — Validation logic (group/attribute-sync overlap checks, extracting group names from group_config).
  • slack_handler_lambda.tf / perm_revoker_lambda.tf — Lambda function definitions. Both read config from the same S3 object.

Python Layer (src/)

Config & Models:

  • config.py — Loads config from S3 (load_approval_config_from_s3), parses into Statement/GroupStatement via parse_statement/parse_group_statement. Config is a Pydantic BaseSettings singleton with frozenset fields for statements, accounts, groups, permission_sets.
  • statement.py — BaseStatement (has permission_set, required_group_membership, etc.) → Statement (account access). GroupStatement is separate (does not inherit BaseStatement, lacks permission_set and required_group_membership). Eligibility filtering (get_eligible_statements_for_user) currently only applies to Statement, not GroupStatement.
  • events.py — RevokeEvent/GroupRevokeEvent and their Scheduled* wrappers. Tagged union via action field literal types.
  • entities/ — Pydantic models for AWS resources (Account, PermissionSet, SSOGroup) and Slack users.

Request Flow (Slack → Decision → Execution):

  • main.py — Slack bolt app. Handles account access shortcuts, modal interactions, button clicks. Registers handlers via app.shortcut(), app.view(), app.action(). The handle_button_click function falls through to group.handle_group_button_click on parse failure.
  • group.py — Parallel handlers for group access shortcuts, submission, and button clicks.
  • slack_helpers.py — View builders (RequestForAccessView, RequestForGroupAccessView), payload parsers (ButtonClickedPayload, ButtonGroupClickedPayload), message block builders. Modal uses dynamic view updates: accounts load first, then permission sets update on account selection via dispatch_action=True.
  • access_control.py — Decision logic (make_decision_on_access_request, make_decision_on_approve_request) is already unified for both Statement and GroupStatement. Execution is split: execute_decision (account) vs execute_decision_on_group_request (group).

Revocation & Scheduling:

  • schedule.py — schedule_revoke_event/schedule_group_revoke_event create EventBridge schedules that fire the revoker Lambda.
  • revoker.py — Handles scheduled revocations, early revocations, inconsistency checks, extend-grant button posting. Parallel functions for account and group flows.

SSO Operations:

  • sso.py — AWS SSO/Identity Store API calls. Account operations (create/delete assignment) and group operations (add/remove membership) are fundamentally different APIs. User lookup (get_user_principal_id_by_email) is shared.

Attribute Sync (separate feature):

  • attribute_syncer.py, attribute_mapper.py, sync_state.py, sync_config.py, sync_notifications.py — Automatic user-to-group sync based on Identity Store attributes. Runs on its own Lambda/schedule.

Key Design Patterns

  • Config uses two separate Terraform variables (config and group_config) that map directly to statements and group_statements in the S3 JSON. Python reads these as separate lists.
  • Pydantic models use frozen=True (ConfigDict(frozen=True) in entities/model.py). Extra fields are silently ignored (default Pydantic v2 behavior, no extra="forbid").
  • parse_group_statement manually extracts keys from the raw dict (doesn't pass the whole dict to Pydantic), so extra keys like ResourceType in the S3 JSON are harmlessly ignored.
  • Decision logic is unified in access_control.py — same make_decision_on_access_request handles both FrozenSet[Statement] and FrozenSet[GroupStatement].
  • Execution logic is deliberately split — account and group execution have different SSO API calls, audit fields, and scheduling.
  • Tests use Hypothesis (src/tests/strategies.py) for property-based testing of statement parsing alongside standard pytest.