Files
DanConwayDev 9fa726bbe1 fix(audit): scope state to the selected repository's maintainers
Fetching kind 30618 by identifier alone mixed independently owned repositories. A newer unrelated state could supply the wrong expected refs even though both repositories were valid on the relay.

Fetch announcements alongside state, select each author's NIP-01-preferred announcement, and resolve current publishers from the selected coordinate. Support reciprocal legacy/indexed roles, transitive confirmation, departures and explicit lead forwarding; reject unresolved lead paths. Verify and scope events before selecting the newest authorized state. Keep the audit crate independent of ngit-grasp.

Authority uses the probed relay's visible announcements. This does not change server authorization, add cross-relay discovery or provide an atomic Git/Nostr snapshot.

Validation: 95 grasp-audit tests pass (5 ignored), including membership and ordering regressions. The local-relay integration target passes all 55 tests, with an end-to-end same-identifier collision and injected stale refs. Workspace all-target Clippy and formatting pass.

Assisted-by: GPT-6
2026-09-17 16:32:16 +00:00

479 lines
20 KiB
Markdown

# GRASP Audit
A reusable audit and compliance testing tool for GRASP protocol implementations.
## Features
- ✅ **Shared Fixtures**: Fixtures cached and reused across tests (default for CLI) to reduce rate-limitting
- ✅ **Isolated Testing**: Fresh fixtures per test for parallel test isolation
- ✅ **Clean Audit Events**: Special tags for easy cleanup (no deletion trails)
- ✅ **Spec-Mirrored Tests**: Test structure matches GRASP protocol exactly
- ✅ **Reusable**: Can test any GRASP implementation (Rust, Go, Python, etc.)
## Quick Start
```bash
# Install
cd grasp-audit
cargo install --path .
# Probe a relay (read-only health check — no keys needed)
grasp-audit probe --relay wss://relay.ngit.dev
# Full compliance audit
grasp-audit audit --relay wss://relay.ngit.dev
```
## Usage Examples
### Probe (Health Check)
A fast, lightweight smoke test that checks whether a relay is healthy. Read-only by default — no keys or write access required.
```bash
# Basic health check
grasp-audit probe --relay wss://relay.ngit.dev
# Machine-readable JSON (one line per run — pipe-friendly)
grasp-audit probe --relay wss://relay.ngit.dev --json
# Continuous monitoring (re-runs every 10 seconds)
grasp-audit probe --relay wss://relay.ngit.dev --watch 10
# Pipe JSON output to a log file
grasp-audit probe --relay wss://relay.ngit.dev --json --watch 10 >> probe.log
# Harden a scheduled read-only probe against untrusted network targets
grasp-audit probe --relay wss://relay.ngit.dev --json --harden-network
# Full write-path check: publish events, git push, verify refs match state
# (requires write access; use --nsec for whitelisted relays)
grasp-audit probe --relay wss://relay.ngit.dev --create-repo
grasp-audit probe --relay wss://relay.ngit.dev --create-repo --nsec nsec1...
```
`--harden-network` is opt-in so `ws://localhost` remains available for local
development. Scheduled probes of discovered servers should enable it. Hardened
probes require `wss://`, pin validated public addresses for WebSocket and HTTP
connections, refuse redirects and cross-origin clone URLs, and bound response
bodies. They are deliberately read-only and cannot be combined with
`--create-repo`.
**Probe checks (read-only):**
| Check | What it verifies |
|---|---|
| `connect_websocket` | WebSocket connection succeeds |
| `nip11_fetch` | NIP-11 relay info document is served (shows software & version) |
| `serves_latest_announcement` | At least one kind:30617 repo announcement is served |
| `git_fetch_refs` | Git HTTP info/refs endpoint responds |
| `git_refs_match_state` | Git refs match the newest kind:30618 state snapshot (lowest event ID breaks timestamp ties) |
The read-only ref check resolves the selected announcement's confirmed maintainer
component before choosing state. Same-identifier repositories owned by unrelated
authors cannot supply its expected refs. Legacy reciprocal listings and indexed
roles are supported; missing or conflicting lead paths fail the check. The probe
uses the announcements visible on the queried relay at the time of the check.
Both read-only and write probes compare the complete branch/tag set: missing,
changed, and unexpected refs fail. HEAD, Nostr refs, and peeled annotated-tag
entries are excluded from that set.
**Additional checks with `--create-repo`:**
| Check | What it verifies |
|---|---|
| `publish_events` | Relay accepts kind:30617 and kind:30618 events |
| `git_repo_initialised` | Relay initialises the git repo after events are published |
| `git_push` | Git push succeeds |
**Overall timeout:** The probe enforces an overall deadline of `min(20s, --watch interval)` to prevent overlapping runs. If the deadline fires, completed checks retain their real results; the step that couldn't start is marked failed with a diagnosis (single slow check vs. cumulative slowness); remaining checks are marked skipped.
### Audit (Compliance Tests)
```bash
# Audit a production GRASP relay (shared fixtures - default)
grasp-audit audit --relay wss://relay.ngit.dev
# Audit local development relay
grasp-audit audit --relay ws://localhost:7334 --spec nip01-smoke
# Run with isolated fixtures (for testing/debugging)
grasp-audit audit --relay ws://localhost:7334 --mode isolated --spec push-auth
# Emit one stable JSON document for a scheduled full audit
grasp-audit audit --relay wss://relay.ngit.dev --spec all --json
```
`audit --json` writes one schema-versioned document to stdout and suppresses
the human progress/report output. Completed runs include tool and run metadata,
pinned GRASP revisions, aggregate counts, and per-test outcomes. Invalid
arguments and setup/runtime errors use the same envelope with an `error`
outcome. Exit status is `0` for a passing audit, `1` for completed test
failures, and `2` for invocation or runtime errors.
### Audit identity
Probe write checks and full audits can use an explicit identity. Key sources
are checked in this order:
1. `--nsec <nsec-or-hex>`
2. `--nsec-file <path>`
3. `GRASP_AUDIT_NSEC`
`--nsec` is convenient for disposable keys, but command arguments may be
visible to other local users. Prefer `--nsec-file` or `GRASP_AUDIT_NSEC` for
scheduled jobs. The two command-line options cannot be used together. If no
source is supplied, the tool generates a fresh key as before.
```bash
grasp-audit probe --relay wss://relay.ngit.dev --create-repo --nsec nsec1...
grasp-audit audit --relay wss://relay.ngit.dev --nsec-file /run/credentials/audit_nsec
GRASP_AUDIT_NSEC=nsec1... grasp-audit audit --relay wss://relay.ngit.dev
```
### As a Library (Audit)
```rust
use grasp_audit::*;
#[tokio::main]
async fn main() -> Result<()> {
// Create audit client with isolated fixtures (recommended for library use)
let config = AuditConfig::isolated();
// let config = AuditConfig::shared(); // Alternative: shared fixtures
let client = AuditClient::new("ws://localhost:7334", config).await?;
// Run NIP-01 smoke tests
let results = specs::Nip01SmokeTests::run_all(&client).await;
results.print_report();
if !results.all_passed() {
std::process::exit(1);
}
Ok(())
}
```
## Test Specifications
The audit tool provides **good test coverage of GRASP-01** requirements, with additional smoke tests for basic Nostr relay functionality and git over HTTP.
### GRASP-01 Tests
Test coverage of GRASP-01 specification:
- Repository announcement acceptance
- State event handling
- Push authorization (selected coordinate, direct maintainer, and transitively confirmed maintainer)
- Event acceptance policy
- Git clone over HTTP
- CORS headers
- NIP-11 relay information document
### NIP-01 Smoke Tests (6 tests)
Basic Nostr relay functionality validation:
1. `websocket_connection` - Can connect to /
2. `send_receive_event` - Can send EVENT, get OK
3. `create_subscription` - Can subscribe with REQ
4. `close_subscription` - Can close subscriptions
5. `reject_invalid_signature` - Rejects bad signatures
6. `reject_invalid_event_id` - Rejects wrong IDs
**Why only smoke tests?** rust-nostr already has 1000+ tests for NIP-01 compliance. We focus on GRASP-specific behavior.
### Git over HTTP Smoke Tests
Basic validation that git clone works over HTTP.
## Fixture Modes
The audit tool supports two fixture caching modes that control how test prerequisites are managed. This is a key feature for controlling test isolation and resource efficiency.
### Shared Mode (Default for CLI)
**Default for CLI usage.** Fixtures are cached and reused across all tests for efficiency.
Use this when:
- Auditing production or development relays
```bash
# CLI uses shared mode by default
grasp-audit audit --relay wss://relay.ngit.dev
```
```rust
let config = AuditConfig::shared();
```
### Isolated Mode (Recommended for Library)
**Recommended for library/test usage.** Each test creates fresh fixtures for complete isolation.
Use this when:
- Using grasp-audit as a library
- Running `cargo test` in parallel
- Tests must not interfere with each other
- Debugging test failures
```bash
# Use isolated mode explicitly
grasp-audit audit --relay ws://localhost:7334 --mode isolated
```
```rust
let config = AuditConfig::isolated();
```
### When to Use Each Mode
| Scenario | Recommended Mode |
| ----------------------------- | ---------------- |
| CLI auditing production relay | Shared (default) |
| CLI auditing local relay | Shared (default) |
| Library usage / `cargo test` | Isolated |
| CI/CD pipeline | Isolated |
| Debugging a single test | Isolated |
## Audit Event Strategy
All audit events automatically include special tags for isolation and cleanup:
```json
{
"tags": [
["t", "grasp-audit-test-event"],
["t", "audit-ci-a1b2c3d4-e5f6-7890-abcd-ef1234567890"],
["t", "audit-cleanup-after-1730822334"]
]
}
```
**Tag Format:**
- `["t", "grasp-audit-test-event"]` - Identifies all audit-related events
- `["t", "audit-{run_id}"]` - Unique identifier for each audit run
- Shared mode: `audit-audit-{uuid}`
- Isolated mode: `audit-isolated-{uuid}`
- `["t", "audit-cleanup-after-{unix_timestamp}"]` - Cleanup scheduling
- Default: Current time + 3600 seconds (1 hour)
**Benefits:**
- **Automatic**: Tags added automatically to all events via `AuditEventBuilder`
- **Isolation**: Each test run has unique ID for event filtering
- **Cleanup**: Events marked for cleanup after timestamp (direct database cleanup)
- **No deletion trails**: No NIP-09 deletion events needed
- **Discovery**: Easy to query all audit events via hashtag
## Architecture
```
grasp-audit/
├── src/
│ ├── lib.rs # Public API
│ ├── audit.rs # Audit config and event tagging
│ ├── client.rs # Audit client
│ ├── fixtures.rs # TestContext and FixtureKind
│ ├── result.rs # Test result types
│ ├── isolation.rs # Test isolation utilities
│ └── specs/
│ ├── mod.rs
│ ├── nip01_smoke.rs # NIP-01 smoke tests
│ └── grasp01/ # GRASP-01 compliance tests
└── bin/
└── grasp-audit.rs # CLI tool
```
## Roadmap
Planned features and improvements:
### Near-term
- [ ] **Configurable backoffs for rate limiting** - Allow configuring retry delays when relays rate-limit requests
- [ ] **Delete events per pubkey** - Send NIP-09 deletion events grouped by pubkey for better cleanup on relays that support it
- [ ] **Delete event handling** - Respect NIP-09 support flagged in NIP-11 relay information document
### Future
- [ ] **GRASP-05 support** - Add test coverage for GRASP-05 specification
### Out of Scope
- **GRASP-02 (Proactive Sync)** - Testing proactive synchronization behavior is inherently difficult due to its asynchronous nature and reliance on external state. This specification is out of scope for automated compliance testing.
## Development
This section covers patterns and guidelines for contributing new audit tests.
### Test Design Pattern: Fixture-First
To prevent rate-limiting from production relays during testing, we use a **fixture-first** approach that minimizes relay interactions.
#### Quick Start for New Tests
1. Create TestContext at test start
2. Get prerequisites via `ctx.get_fixture(FixtureKind::...)`
3. Build test-specific events using fixtures as base
4. Verify outcomes via `send_and_verify_accepted/rejected`
#### Pattern Template
```rust
pub async fn test_something(client: &AuditClient) -> TestResult {
TestResult::new(...)
.run(|| async {
// 1. Context
let ctx = TestContext::new(client);
// 2. Prerequisites (cached per-TestContext)
let repo = ctx.get_fixture(FixtureKind::ValidRepoSent).await?;
// 3. Test-specific event
let my_event = client.create_issue(&repo, "Title", "Content", vec![])?;
// 4. Verify
send_and_verify_accepted(client, my_event, "description").await?;
Ok(())
})
.await
}
```
#### Three-Layer Architecture
```
┌─────────────────────────────────────────────────────────────────┐
│ Layer 3: Test Functions │
│ Create TestContext, get fixtures, build scenarios, verify │
├─────────────────────────────────────────────────────────────────┤
│ Layer 2: FixtureKind + TestContext │
│ ValidRepo, RepoState, OwnerStateDataPushed, etc. │
│ Mode-aware caching within TestContext │
├─────────────────────────────────────────────────────────────────┤
│ Layer 1: AuditClient │
│ event_builder, create_repo_announcement, send_event │
└─────────────────────────────────────────────────────────────────┘
```
#### Available Fixtures
| FixtureKind | Provides | Use When |
| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------ |
| `ValidRepo` | Accepted repo announcement (kind 30617). Signed by owner keys, lists maintainer in maintainers tag. | Need a repo as prerequisite |
| `RepoWithIssue` | Repo + accepted issue (kind 1621) | Testing issue-dependent events |
| `RepoWithComment` | Repo + issue + comment (kind 1111) | Testing comment-dependent events |
| `RepoState` | Repo + state event (kind 30618). Signed by owner, points to `DETERMINISTIC_COMMIT_HASH`. | Testing owner state events |
| `PREvent` | Repo + PR event (kind 1618). Signed by PR author, points to `PR_TEST_COMMIT_HASH`. | Testing PR-dependent events |
| `PREventGenerated` | PR event built but NOT sent to relay. | Need PR event ID before publishing |
| `PRWrongCommitPushedBeforeEvent` | Wrong commit pushed to `refs/nostr/<pr-event-id>` before PR event sent. Returns unsent PR event. | Testing pre-event ref cleanup |
| `PREventSentAfterWrongPush` | PR event sent after wrong commit was pushed. Tests cleanup behavior. | Testing post-event ref cleanup |
| `OwnerStateDataPushed` | Full owner push flow: state event + git data pushed. Points to `DETERMINISTIC_COMMIT_HASH`. | Testing owner push authorization |
| `MaintainerStateDataPushed` | Full maintainer push flow: force-pushes over owner's data. Points to `MAINTAINER_DETERMINISTIC_COMMIT_HASH`. | Testing maintainer push authorization |
| `RecursiveMaintainerStateDataPushed` | Full transitive reciprocal-confirmation flow: selected coordinate → maintainer → additional maintainer. Points to `RECURSIVE_MAINTAINER_DETERMINISTIC_COMMIT_HASH`. | Testing transitively confirmed authorization |
| `HeadSetToDevelopBranch` | State event with HEAD=refs/heads/develop. Depends on RecursiveMaintainerStateDataPushed. | Testing HEAD branch switching |
#### Deterministic Commit Hashes
Fixtures use deterministic commit hashes for reproducible testing:
| Constant | Hash | Used By |
| ------------------------------------------------ | ------------------------------------------ | ------------------------------------------------ |
| `DETERMINISTIC_COMMIT_HASH` | `d6e4b26ccf9c268d18d60e6d09804313cc850821` | Owner fixtures (RepoState, OwnerStateDataPushed) |
| `MAINTAINER_DETERMINISTIC_COMMIT_HASH` | `d26703c007eff6d17fee3bb70ce8be5d1427d0e7` | MaintainerStateDataPushed |
| `RECURSIVE_MAINTAINER_DETERMINISTIC_COMMIT_HASH` | `54a2b4b3cbc3373ad1438b8ffad1681d12bc6c4a` | RecursiveMaintainerStateDataPushed |
| `PR_TEST_COMMIT_HASH` | `5a51b30e4615b572dcd5b9e487861b58605a5c21` | PR fixtures (PREvent, PREventGenerated) |
#### Fixture Dependencies
Fixtures automatically resolve their dependencies:
```
ValidRepo (base)
├── RepoWithIssue → RepoWithComment
├── RepoState
├── PREventGenerated → PRWrongCommitPushedBeforeEvent → PREventSentAfterWrongPush
├── PREvent
└── OwnerStateDataPushed
└── MaintainerStateDataPushed
└── RecursiveMaintainerStateDataPushed
└── HeadSetToDevelopBranch
```
#### Fixture Lifecycle: Generate → Send → Verify → DataPushed
Every fixture follows a lifecycle (some stop earlier):
1. **GENERATE**: Build event via `AuditClient.event_builder()` (in memory only)
2. **SEND**: `client.send_event(event)` transmits to relay (rate-limited operation)
3. **VERIFY**: Query relay to confirm acceptance/rejection
4. **DATA_PUSHED**: (DataPushed variants only) Clone repo, create commit, push to git server
Caching happens after the fixture completes - same fixture request returns cached Event.
**Note:** Some fixtures handle their own event sending (e.g., `OwnerStateDataPushed`, `MaintainerStateDataPushed`). These are marked with `sends_own_events() -> true`.
#### How TestContext Correlates Events
Each TestContext shares a `run_id` with all events:
```rust
// All events in a TestContext get these tags automatically:
["t", "grasp-audit-test-event"] // Identifies test events
["t", "audit-{run_id}"] // Unique ID for this run
["t", "audit-cleanup-after-{ts}"] // Cleanup timestamp
```
This enables:
- Event correlation within a test run
- Production relay cleanup scripts
- Test isolation between runs
#### When NOT to Use Fixtures
Use direct event building (NOT fixtures) when:
- **Testing event REJECTION** - Build invalid events directly
- **Testing signature/ID validation** - Need malformed events
- **One-off connectivity tests** - No prerequisites needed
```rust
// Example: Testing rejection (build invalid event directly)
let invalid_event = client.event_builder(Kind::GitRepoAnnouncement, "")
.tag(Tag::identifier("test"))
// Missing required 'clone' tag - should be rejected
.build(client.keys())?;
send_and_verify_rejected(client, invalid_event, "missing clone tag").await?;
```
#### Anti-Patterns to Avoid
❌ **Creating TestContext inside helper functions** - Tests lose cache control
❌ **Monolithic setup functions** - Mix fixture retrieval with git operations
❌ **Direct event creation when fixture exists** - Misses caching opportunity
✅ **Each test creates own TestContext** - Isolation guaranteed
✅ **Use fixtures for prerequisites** - Caching minimizes relay calls
✅ **Build invalid events directly** - Only for rejection tests
## Contributing
This tool is designed to be reusable by any GRASP implementation. Contributions welcome!
## License
MIT