shellfirm 0.3.7

//! Manage command checks

use std::sync::OnceLock;

use crate::error::Result;
use regex::Regex;
use serde_derive::{Deserialize, Serialize};
use serde_regex;
use tracing::{debug, warn};

use crate::{
    blast_radius::{self, BlastRadiusInfo, BlastScope},
    config::{Challenge, Settings},
    context::{self, RuntimeContext},
    env::Environment,
    policy::{self, MergedPolicy},
    prompt::{AlternativeInfo, ChallengeResult, DisplayContext, Prompter},
};

/// String with all checks from `checks` folder (prepared in build.rs) in YAML
/// format.
const ALL_CHECKS: &str = include_str!(concat!(env!("OUT_DIR"), "/all-checks.yaml"));

/// Severity level of a check — determines how critical a matched pattern is.
///
/// The natural ordering (`Info < Low < Medium < High < Critical`) is used
/// for filtering: when a `min_severity` is configured, only checks at or
/// above that threshold trigger a challenge.
#[derive(Debug, Default, Deserialize, Serialize, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
pub enum Severity {
    Info,
    Low,
    #[default]
    Medium,
    High,
    Critical,
}

impl std::fmt::Display for Severity {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::Info => write!(f, "INFO"),
            Self::Low => write!(f, "LOW"),
            Self::Medium => write!(f, "MEDIUM"),
            Self::High => write!(f, "HIGH"),
            Self::Critical => write!(f, "CRITICAL"),
        }
    }
}

/// A single post-match filter that gates whether a check should fire.
///
/// Filters are evaluated after the regex matches. **All** filters in the
/// list must pass (logical AND) for the check to be kept.
///
/// YAML format (adjacently tagged):
/// ```yaml
/// filters:
///   - type: PathExists
///     value: 1
///   - type: NotContains
///     value: "--dry-run"
/// ```
#[derive(Debug, Deserialize, Serialize, Clone)]
#[serde(tag = "type", content = "value")]
pub enum Filter {
    /// Keep the check only if the captured path exists on disk.
    /// The value is the regex capture-group index (1-based).
    PathExists(usize),
    /// Keep the check only if the command does **not** contain this substring.
    NotContains(String),
    /// Keep the check only if the command **does** contain this substring.
    Contains(String),
}

/// Describe single check
#[derive(Debug, Deserialize, Serialize, Clone)]
pub struct Check {
    pub id: String,
    /// test is a value that we check the command.
    #[serde(with = "serde_regex")]
    pub test: Regex,
    /// description of what is risky in this command
    pub description: String,
    /// the group of the check see files in `checks` folder
    pub from: String,
    #[serde(default)]
    pub challenge: Challenge,
    #[serde(default)]
    pub filters: Vec<Filter>,
    /// Safer command alternative suggestion (shown to user).
    #[serde(default)]
    pub alternative: Option<String>,
    /// Explanation of why the alternative is safer.
    #[serde(default)]
    pub alternative_info: Option<String>,
    /// Severity level — determines how critical this check is.
    #[serde(default)]
    pub severity: Severity,
}

/// Return a cached reference to all built-in check patterns.
///
/// The YAML is parsed and regexes are compiled exactly once (on first call).
/// Subsequent calls return a reference to the cached static slice.
pub(crate) fn all_checks_cached() -> &'static [Check] {
    static CHECKS: OnceLock<Vec<Check>> = OnceLock::new();
    CHECKS.get_or_init(|| serde_yaml::from_str(ALL_CHECKS).expect("built-in checks are valid YAML"))
}

/// Return all built-in shellfirm check patterns
///
/// # Errors
/// when has an error when parsing check str to [`Check`] list
pub fn get_all() -> Result<Vec<Check>> {
    Ok(all_checks_cached().to_vec())
}

/// Load custom checks from YAML files in a directory.
///
/// # Errors
/// When a file cannot be read or parsed.
pub fn load_custom_checks(checks_dir: &std::path::Path) -> Result<Vec<Check>> {
    let mut custom_checks = Vec::new();
    if !checks_dir.is_dir() {
        return Ok(custom_checks);
    }
    let entries = std::fs::read_dir(checks_dir)?;
    for entry in entries {
        let entry = entry?;
        let path = entry.path();
        if path.extension().is_some_and(|e| e == "yaml" || e == "yml") {
            let content = std::fs::read_to_string(&path)?;
            let checks: Vec<Check> = serde_yaml::from_str(&content)?;
            custom_checks.extend(checks);
        }
    }
    Ok(custom_checks)
}

/// Validate check definitions and return a list of warning messages.
///
/// Currently checks:
/// - `PathExists(n)` — warns when `n` exceeds the number of capture groups
///   in the check's regex.
#[must_use]
pub fn validate_checks(checks: &[Check]) -> Vec<String> {
    let mut warnings = Vec::new();
    for check in checks {
        let num_captures = check.test.captures_len(); // group 0 + real groups
        for filter in &check.filters {
            if let Filter::PathExists(group_idx) = filter {
                if *group_idx >= num_captures {
                    warnings.push(format!(
                        "check {:?}: PathExists({}) references a capture group that \
                         does not exist (regex has {} groups including group 0)",
                        check.id, group_idx, num_captures
                    ));
                }
            }
        }
    }
    warnings
}

/// New challenge flow that is context-aware and uses the Prompter trait.
///
/// This is the primary entry point for the v1.0 pipeline.
/// The effective challenge is computed by applying six escalation layers
/// in order — each layer can only make the challenge stricter, never weaker:
///
/// 1. **Base** — `settings.challenge`
/// 2. **Severity** — `severity_escalation` (Critical→Yes, High→Enter by default)
/// 3. **Group** — `group_escalation[check.from]`
/// 4. **Check-ID** — `check_escalation[check.id]`
/// 5. **Context** — `escalate_challenge(risk_level)` (SSH→Enter, root/prod→Yes)
/// 6. **Policy** — `merged_policy.effective_challenge(check.id)` (.shellfirm.yaml)
///
/// # Errors
/// Returns an error if the prompter fails.
pub fn challenge_with_context(
    settings: &Settings,
    checks: &[&Check],
    context: &RuntimeContext,
    merged_policy: &MergedPolicy,
    prompter: &dyn Prompter,
    blast_radii: &[(String, BlastRadiusInfo)],
) -> Result<ChallengeResult> {
    let mut descriptions: Vec<String> = Vec::new();
    let mut alternatives: Vec<AlternativeInfo> = Vec::new();
    let mut should_deny_command = false;

    debug!(
        "list of denied pattern ids {:?}",
        settings.deny_patterns_ids
    );

    for check in checks {
        if !descriptions.contains(&check.description) {
            descriptions.push(check.description.clone());
        }
        // Check deny from global settings
        if !should_deny_command && settings.deny_patterns_ids.contains(&check.id) {
            should_deny_command = true;
        }
        // Check deny from project policy
        if !should_deny_command && merged_policy.is_denied(&check.id) {
            should_deny_command = true;
        }
        // Collect alternatives
        if let Some(ref alt) = check.alternative {
            let already_has = alternatives.iter().any(|a| a.suggestion == *alt);
            if !already_has {
                alternatives.push(AlternativeInfo {
                    suggestion: alt.clone(),
                    explanation: check.alternative_info.clone(),
                });
            }
        }
    }

    // --- 6-layer escalation pipeline ---
    let base_challenge = settings.challenge;

    // Layer 1: base challenge
    let mut effective = base_challenge;

    // Layer 2: severity escalation (highest severity across all matched checks)
    let max_severity = checks.iter().map(|c| c.severity).max();
    if let Some(sev) = max_severity {
        if let Some(sev_floor) = settings.severity_escalation.challenge_for_severity(sev) {
            effective = max_challenge(effective, sev_floor);
        }
    }

    // Layer 3: group escalation (per-group floor from settings)
    for check in checks {
        if let Some(&group_floor) = settings.group_escalation.get(&check.from) {
            effective = max_challenge(effective, group_floor);
        }
    }

    // Layer 4: check-ID escalation (per-check floor from settings)
    for check in checks {
        if let Some(&check_floor) = settings.check_escalation.get(&check.id) {
            effective = max_challenge(effective, check_floor);
        }
    }

    // Layer 5: context escalation (SSH→Enter, root/prod→Yes)
    effective = max_challenge(
        effective,
        context::escalate_challenge(
            &base_challenge,
            context.risk_level,
            &settings.context.escalation,
        ),
    );

    // Layer 6: project policy overrides (.shellfirm.yaml)
    for check in checks {
        let policy_effective = merged_policy.effective_challenge(&check.id, &effective);
        effective = max_challenge(effective, policy_effective);
    }

    // Build escalation note
    let escalation_note = if effective == base_challenge {
        None
    } else {
        Some(format!("{base_challenge} -> {effective}"))
    };

    // Compute highest severity for display
    let severity_label = max_severity.map(|s| format!("{s}"));

    // Build blast radius label from the highest-scope entry
    let blast_radius_label = blast_radii
        .iter()
        .max_by_key(|(_, br)| br.scope)
        .map(|(_, br)| format!("[{}] — {}", br.scope, br.description));

    let display = DisplayContext {
        is_denied: should_deny_command,
        descriptions,
        alternatives,
        context_labels: context.labels.clone(),
        effective_challenge: effective,
        escalation_note,
        severity_label,
        blast_radius_label,
    };

    Ok(prompter.run_challenge(&display))
}

/// Check if the given command matched to one of the checks.
///
/// Uses [`crate::env::RealEnvironment`] for filter evaluation (filesystem checks).
/// For testing, prefer [`run_check_on_command_with_env`] with a mock environment.
///
/// # Arguments
///
/// * `checks` - List of checks that we want to validate.
/// * `command` - Command check.
#[must_use]
pub fn run_check_on_command<'a>(checks: &'a [Check], command: &str) -> Vec<&'a Check> {
    run_check_on_command_with_env(checks, command, &crate::env::RealEnvironment)
}

/// Like `run_check_on_command` but uses the Environment trait for filters.
#[must_use]
pub fn run_check_on_command_with_env<'a>(
    checks: &'a [Check],
    command: &str,
    env: &dyn Environment,
) -> Vec<&'a Check> {
    checks
        .iter()
        .filter(|v| v.test.is_match(command))
        .filter(|v| check_custom_filter_with_env(v, command, env))
        .collect()
}

/// Evaluate filters using the [`Environment`] trait (testable version).
///
/// Returns `true` when the check should be kept (all filters pass).
fn check_custom_filter_with_env(check: &Check, command: &str, env: &dyn Environment) -> bool {
    if check.filters.is_empty() {
        return true;
    }
    let caps = check.test.captures(command);

    for filter in &check.filters {
        debug!("filter information: command {command:?} filter: {filter:?}");

        let keep = match filter {
            Filter::PathExists(group_idx) => {
                let file_path = caps
                    .as_ref()
                    .and_then(|c| c.get(*group_idx))
                    .map_or_else(
                        || {
                            warn!(
                                "check {:?}: PathExists references capture group {} which does not exist in regex",
                                check.id, group_idx
                            );
                            ""
                        },
                        |m| m.as_str(),
                    );
                if file_path.is_empty() {
                    // No path captured → treat as "path doesn't exist" → suppress check
                    false
                } else {
                    filter_path_exists_with_env(file_path, env)
                }
            }
            Filter::NotContains(ref s) => !command.contains(s.as_str()),
            Filter::Contains(ref s) => command.contains(s.as_str()),
        };

        if !keep {
            return false;
        }
    }

    true
}

/// Check if path exists using [`Environment`] trait.
fn filter_path_exists_with_env(file_path: &str, env: &dyn Environment) -> bool {
    use std::borrow::Cow;

    let trimmed = file_path.trim();
    let file_path: Cow<'_, str> = if trimmed.starts_with('~') {
        match env.home_dir() {
            Some(home) => Cow::Owned(trimmed.replacen('~', &home.display().to_string(), 1)),
            None => return true,
        }
    } else {
        Cow::Borrowed(trimmed)
    };

    if file_path.contains('*') {
        return true;
    }

    let full_path = match env.current_dir() {
        Ok(cwd) => cwd.join(&*file_path),
        Err(err) => {
            debug!("could not get current dir. err: {err:?}");
            return true;
        }
    };

    debug!("check if {} path exists", full_path.display());
    env.path_exists(&full_path)
}

/// Return the stricter of two challenges.
pub(crate) fn max_challenge(a: Challenge, b: Challenge) -> Challenge {
    a.stricter(b)
}

/// Merge two vectors of check references, deduplicating by check ID.
/// Primary matches are kept first; secondary matches are appended only if
/// their ID has not been seen yet.
fn dedup_check_matches<'a>(primary: Vec<&'a Check>, secondary: Vec<&'a Check>) -> Vec<&'a Check> {
    let mut seen_ids = std::collections::HashSet::new();
    let mut result = Vec::with_capacity(primary.len() + secondary.len());
    for m in primary.into_iter().chain(secondary) {
        if seen_ids.insert(m.id.as_str()) {
            result.push(m);
        }
    }
    result
}

/// The result of running the core analysis pipeline on a command.
///
/// This is the shared data structure consumed by both the terminal prompter
/// (shell hooks) and the MCP/agent JSON response path.
#[derive(Debug, Clone)]
pub struct PipelineResult {
    /// The original command after quote-stripping.
    pub stripped_command: String,
    /// The subcommand parts after splitting on shell operators.
    pub command_parts: Vec<String>,
    /// Checks that matched and are at or above `min_severity`.
    pub active_matches: Vec<Check>,
    /// Checks that matched but are below `min_severity`.
    pub skipped_matches: Vec<Check>,
    /// The detected runtime context (full — for audit logging).
    pub context: RuntimeContext,
    /// Context filtered to only signals relevant to matched check groups
    /// (for display and challenge escalation).
    pub relevant_context: RuntimeContext,
    /// The highest severity across all matches (active + skipped).
    pub max_severity: Severity,
    /// Whether any matched check is on the deny list.
    pub is_denied: bool,
    /// Safer alternative suggestions collected from matched checks.
    pub alternatives: Vec<AlternativeInfo>,
    /// The merged project policy (if any).
    pub merged_policy: MergedPolicy,
    /// Runtime-computed blast radius for matched checks (`check_id`, info).
    pub blast_radii: Vec<(String, BlastRadiusInfo)>,
    /// The widest blast scope across all computed blast radii.
    pub max_blast_scope: Option<BlastScope>,
}

/// Run the core shellfirm analysis pipeline on a command string.
///
/// This is the shared entry point used by both the `pre-command` CLI path
/// and the MCP server / agent guardrails path. It performs:
/// 1. Strip quoted strings
/// 2. Split on shell operators
/// 3. Pattern matching (parallelized)
/// 4. Context detection
/// 5. Policy discovery & merging
/// 6. Severity filtering
///
/// The caller is responsible for the presentation layer (terminal challenge
/// or structured JSON response).
///
/// # Errors
/// Returns an error if check loading or policy parsing fails.
#[allow(clippy::too_many_lines)]
pub fn analyze_command(
    command: &str,
    settings: &Settings,
    checks: &[Check],
    env: &dyn Environment,
    strip_quotes_re: &Regex,
) -> Result<PipelineResult> {
    // 1. Strip quoted strings
    let stripped = strip_quotes_re.replace_all(command, "").into_owned();

    // 2. Split on shell operators
    let command_parts = split_command(&stripped);

    debug!("analyze_command: parts={command_parts:?}");

    // 3. Pattern matching — each segment independently
    let segment_matches: Vec<&Check> = command_parts
        .iter()
        .flat_map(|c| run_check_on_command_with_env(checks, c, env))
        .collect();

    // Also match against the full stripped command for cross-operator patterns
    // (e.g. "history | bash" where the pipe is normally a split point).
    let matches = if command_parts.len() > 1 {
        let full_matches = run_check_on_command_with_env(checks, &stripped, env);
        dedup_check_matches(segment_matches, full_matches)
    } else {
        segment_matches
    };

    debug!("analyze_command: {} matches found", matches.len());

    // 4. Policy discovery (filesystem walk only — no subprocess spawns)
    let cwd = env.current_dir().unwrap_or_default();
    let project_policy = policy::discover(env, &cwd);

    // Check if policy adds extra checks that might match
    let policy_has_extra_checks = project_policy
        .as_ref()
        .is_some_and(|pp| !pp.checks.is_empty());

    // FAST PATH: no matches and no policy extra checks → skip context detection
    // entirely. This avoids spawning git/kubectl subprocesses for the common case
    // where a command (e.g. "ls", "cd", "cargo build") matches nothing.
    if matches.is_empty() && !policy_has_extra_checks {
        debug!("analyze_command: no matches and no policy checks, skipping context detection");
        return Ok(PipelineResult {
            stripped_command: stripped,
            command_parts,
            active_matches: Vec::new(),
            skipped_matches: Vec::new(),
            context: RuntimeContext::default(),
            relevant_context: RuntimeContext::default(),
            max_severity: Severity::default(),
            is_denied: false,
            alternatives: Vec::new(),
            merged_policy: MergedPolicy::default(),
            blast_radii: Vec::new(),
            max_blast_scope: None,
        });
    }

    // 5. Context detection (only when we have matches or policy extra checks)
    let runtime_context = context::detect(env, &settings.context);

    // 6. Policy merging (needs git branch from context for branch-specific overrides)
    let merged_policy = if let Some(ref pp) = project_policy {
        policy::merge_into_settings(settings, pp, runtime_context.git_branch.as_deref())
    } else {
        MergedPolicy::default()
    };

    // Merge extra checks from project policy
    let mut all_matches: Vec<&Check> = matches;
    if !merged_policy.extra_checks.is_empty() {
        let extra_segment: Vec<&Check> = command_parts
            .iter()
            .flat_map(|c| run_check_on_command_with_env(&merged_policy.extra_checks, c, env))
            .collect();
        let extra = if command_parts.len() > 1 {
            let extra_full =
                run_check_on_command_with_env(&merged_policy.extra_checks, &stripped, env);
            dedup_check_matches(extra_segment, extra_full)
        } else {
            extra_segment
        };
        all_matches.extend(extra);
    }

    // 7. Severity filtering
    let (active_refs, skipped_refs): (Vec<&Check>, Vec<&Check>) =
        if let Some(min_sev) = settings.min_severity {
            all_matches.into_iter().partition(|c| c.severity >= min_sev)
        } else {
            (all_matches, Vec::new())
        };

    let active_matches: Vec<Check> = active_refs.into_iter().cloned().collect();
    let skipped_matches: Vec<Check> = skipped_refs.into_iter().cloned().collect();

    // Compute max severity across all matches
    let max_severity = active_matches
        .iter()
        .chain(skipped_matches.iter())
        .map(|c| c.severity)
        .max()
        .unwrap_or_default();

    // Check deny lists
    let is_denied = active_matches
        .iter()
        .any(|c| settings.deny_patterns_ids.contains(&c.id) || merged_policy.is_denied(&c.id));

    // Collect alternatives
    let mut alternatives = Vec::new();
    for check in &active_matches {
        if let Some(ref alt) = check.alternative {
            let already_has = alternatives
                .iter()
                .any(|a: &AlternativeInfo| a.suggestion == *alt);
            if !already_has {
                alternatives.push(AlternativeInfo {
                    suggestion: alt.clone(),
                    explanation: check.alternative_info.clone(),
                });
            }
        }
    }

    // 8. Blast radius computation (only for active matches, when enabled)
    let blast_radii = if settings.blast_radius {
        blast_radius::compute_for_matches(&active_matches, &command_parts, &stripped, env)
    } else {
        Vec::new()
    };
    let max_blast_scope = blast_radii.iter().map(|(_, br)| br.scope).max();

    // Compute context filtered to only signals relevant to matched groups
    let matched_groups: std::collections::HashSet<&str> =
        active_matches.iter().map(|c| c.from.as_str()).collect();
    let relevant_context = runtime_context.filter_for_groups(&matched_groups, &settings.context);

    Ok(PipelineResult {
        stripped_command: stripped,
        command_parts,
        active_matches,
        skipped_matches,
        context: runtime_context,
        relevant_context,
        max_severity,
        is_denied,
        alternatives,
        merged_policy,
        blast_radii,
        max_blast_scope,
    })
}

/// Split a command string into parts by shell operators.
/// Handles `&&`, `||`, `|`, and `;` while respecting single and double quotes.
/// Operators inside quoted strings are not treated as separators.
#[must_use]
pub fn split_command(command: &str) -> Vec<String> {
    let mut parts = Vec::new();
    let bytes = command.as_bytes();
    let len = bytes.len();
    let mut i = 0;
    let mut start = i;
    let mut in_single_quote = false;
    let mut in_double_quote = false;

    while i < len {
        let b = bytes[i];

        // Track quote state (all delimiters are ASCII, safe to scan bytes)
        if b == b'\'' && !in_double_quote {
            in_single_quote = !in_single_quote;
            i += 1;
        } else if b == b'"' && !in_single_quote {
            in_double_quote = !in_double_quote;
            i += 1;
        } else if !in_single_quote && !in_double_quote {
            // Check for two-character operators first
            let (is_split, advance) = if i + 1 < len
                && ((b == b'&' && bytes[i + 1] == b'&') || (b == b'|' && bytes[i + 1] == b'|'))
            {
                (true, 2)
            } else if b == b'|' || b == b';' {
                (true, 1)
            } else {
                (false, 1)
            };

            if is_split {
                parts.push(command[start..i].to_string());
                i += advance;
                start = i;
            } else {
                i += 1;
            }
        } else {
            // Inside quotes — pass through without splitting
            i += 1;
        }
    }
    if start < len {
        parts.push(command[start..].to_string());
    }
    parts
}

#[cfg(test)]
mod test_checks {
    use insta::{assert_debug_snapshot, with_settings};

    use super::*;

    const CHECKS: &str = r###"
- from: test-1
  test: test-(1)
  enable: true
  description: ""
  id: ""
- from: test-2
  test: test-(1|2)
  enable: true
  description: ""
  id: ""
- from: test-disabled
  test: test-disabled
  enable: true
  description: ""
  id: ""
"###;

    #[test]
    fn can_run_check_on_command() {
        let checks: Vec<Check> = serde_yaml::from_str(CHECKS).unwrap();
        with_settings!({filters => vec![
            // Normalize Regex debug format across insta/regex crate versions
            (r#"(?s)test:\s*Regex\(\s*"([^"]+)",?\s*\)"#, "test: $1"),
        ]}, {
            assert_debug_snapshot!(run_check_on_command(&checks, "test-1"));
            assert_debug_snapshot!(run_check_on_command(&checks, "unknown command"));
        });
    }

    #[test]
    fn can_check_custom_filter_with_file_exists() {
        use std::collections::HashSet;
        let filters = vec![Filter::PathExists(1)];

        let check = Check {
            id: "id".to_string(),
            test: Regex::new("(?:^|[^>])>([^>].*)").unwrap(),
            description: "some description".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters,
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        };

        // Use mock environment: file does NOT exist
        let env_no_file = crate::env::MockEnvironment {
            cwd: "/mock".into(),
            ..Default::default()
        };
        let command = "cat 'write message' > /mock/app/message.txt";
        assert_debug_snapshot!(check_custom_filter_with_env(&check, command, &env_no_file));

        // Use mock environment: file DOES exist
        let mut existing = HashSet::new();
        existing.insert(std::path::PathBuf::from("/mock/app/message.txt"));
        let env_with_file = crate::env::MockEnvironment {
            cwd: "/mock".into(),
            existing_paths: existing,
            ..Default::default()
        };
        assert_debug_snapshot!(check_custom_filter_with_env(
            &check,
            command,
            &env_with_file
        ));
    }

    #[test]
    fn can_check_custom_filter_with_str_contains() {
        let filters = vec![Filter::NotContains("--dry-run".to_string())];

        let check = Check {
            id: "id".to_string(),
            test: Regex::new("(delete)").unwrap(),
            description: "some description".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters,
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        };

        let env = crate::env::MockEnvironment::default();
        assert_debug_snapshot!(check_custom_filter_with_env(&check, "delete", &env));
        assert_debug_snapshot!(check_custom_filter_with_env(
            &check,
            "delete --dry-run",
            &env
        ));
    }

    #[test]
    fn can_check_custom_filter_with_contains() {
        let filters = vec![Filter::Contains("--force".to_string())];

        let check = Check {
            id: "id".to_string(),
            test: Regex::new("(push)").unwrap(),
            description: "some description".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters,
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        };

        let env = crate::env::MockEnvironment::default();
        // Without --force the filter suppresses the check
        assert!(!check_custom_filter_with_env(
            &check,
            "git push origin main",
            &env
        ));
        // With --force the filter keeps the check
        assert!(check_custom_filter_with_env(
            &check,
            "git push --force origin main",
            &env
        ));
    }

    #[test]
    fn can_check_custom_filter_with_missing_capture_group() {
        // PathExists references group 5, but regex only has 1 group
        let filters = vec![Filter::PathExists(5)];

        let check = Check {
            id: "test-missing-group".to_string(),
            test: Regex::new("rm (.*)").unwrap(),
            description: "some description".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters,
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        };

        let env = crate::env::MockEnvironment::default();
        // Should return false (suppress check) because group 5 doesn't exist
        assert!(!check_custom_filter_with_env(&check, "rm /tmp/foo", &env));
    }

    #[test]
    fn can_check_multiple_filters_all_must_pass() {
        let filters = vec![
            Filter::NotContains("--dry-run".to_string()),
            Filter::NotContains("--check".to_string()),
        ];

        let check = Check {
            id: "id".to_string(),
            test: Regex::new("(delete)").unwrap(),
            description: "some description".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters,
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        };

        let env = crate::env::MockEnvironment::default();
        // Both absent → check fires
        assert!(check_custom_filter_with_env(&check, "delete", &env));
        // --dry-run present → suppressed
        assert!(!check_custom_filter_with_env(
            &check,
            "delete --dry-run",
            &env
        ));
        // --check present → suppressed
        assert!(!check_custom_filter_with_env(
            &check,
            "delete --check",
            &env
        ));
        // Both present → suppressed
        assert!(!check_custom_filter_with_env(
            &check,
            "delete --dry-run --check",
            &env
        ));
    }

    #[test]
    fn can_get_all_checks() {
        assert_debug_snapshot!(get_all().is_ok());
    }

    #[test]
    fn test_split_command_and_and() {
        let parts = split_command("ls && rm -rf /");
        assert_eq!(parts, vec!["ls ", " rm -rf /"]);
    }

    #[test]
    fn test_split_command_pipe() {
        let parts = split_command("cat foo | grep bar");
        assert_eq!(parts, vec!["cat foo ", " grep bar"]);
    }

    #[test]
    fn test_split_command_mixed() {
        let parts = split_command("a && b || c; d");
        assert_eq!(parts, vec!["a ", " b ", " c", " d"]);
    }

    #[test]
    fn test_split_command_single() {
        let parts = split_command("git push -f");
        assert_eq!(parts, vec!["git push -f"]);
    }

    #[test]
    fn test_split_command_double_quoted_operator() {
        // Operators inside double quotes should NOT cause a split
        let parts = split_command(r#"echo "hello && world""#);
        assert_eq!(parts, vec![r#"echo "hello && world""#]);
    }

    #[test]
    fn test_split_command_single_quoted_pipe() {
        // Pipe inside single quotes should NOT cause a split
        let parts = split_command("echo 'a | b'");
        assert_eq!(parts, vec!["echo 'a | b'"]);
    }

    #[test]
    fn test_split_command_quoted_then_operator() {
        // Quoted section followed by a real operator
        let parts = split_command(r#"echo "safe" && rm -rf /"#);
        assert_eq!(parts, vec![r#"echo "safe" "#, " rm -rf /"]);
    }

    #[test]
    fn test_all_builtin_checks_pass_validation() {
        let checks = get_all().unwrap();
        let warnings = validate_checks(&checks);
        assert!(
            warnings.is_empty(),
            "Built-in checks have validation warnings:\n{}",
            warnings.join("\n")
        );
    }

    #[test]
    fn test_validate_catches_bad_capture_group() {
        let checks = vec![Check {
            id: "bad".to_string(),
            test: Regex::new("rm (.*)").unwrap(),
            description: "test".to_string(),
            from: "test".to_string(),
            challenge: Challenge::default(),
            filters: vec![Filter::PathExists(5)], // only 2 groups (0 + 1)
            alternative: None,
            alternative_info: None,
            severity: Severity::default(),
        }];
        let warnings = validate_checks(&checks);
        assert_eq!(warnings.len(), 1);
        assert!(warnings[0].contains("PathExists(5)"));
    }
}