gitdataai/libs/service/skill/scanner.rs
ZhenYi 4034e98dfb refactor(service): multi-root skill scanner and chat/join_request updates
Skill scanner now walks .claude/skills and .codex/skills directories
separately, adds relative_path/system fields to DiscoveredSkill, and
supports root-level SKILL.md. Update chat context and join request
handling.
2026-05-18 20:43:27 +08:00

677 lines
23 KiB
Rust

//! Repository skill scanner.
//!
//! Scans repositories for SKILL.md files and upserts skill records.
use crate::error::AppError;
use chrono::Utc;
use git2::Repository;
use models::ActiveModelTrait;
use models::projects::project_skill::ActiveModel as SkillActiveModel;
use models::projects::project_skill::Column as C;
use models::projects::project_skill::Entity as SkillEntity;
use models::repos::repo::Model as RepoModel;
use sea_orm::{ColumnTrait, EntityTrait, QueryFilter, Set};
use sha1::Digest;
use std::path::Path;
use uuid::Uuid;
const SKILL_ROOTS: &[(&str, &str)] = &[(".claude/skills", "claude"), (".codex/skills", "codex")];
const ROOT_SKILL_SYSTEM: &str = "root";
fn should_descend_dir(name: &str) -> bool {
name != ".git"
}
/// Skill discovery result from a single repository.
#[derive(Debug)]
pub struct DiscoveredSkill {
/// URL-safe slug derived from the directory name.
pub slug: String,
/// Human-readable name (from frontmatter or slug).
pub name: String,
/// Short description (from frontmatter).
pub description: Option<String>,
/// Raw markdown body after the frontmatter.
pub content: String,
/// Parsed frontmatter as JSON.
pub metadata: serde_json::Value,
/// Relative path of the SKILL.md file in the repository.
pub relative_path: Option<String>,
/// Skill system/source directory, e.g. "claude" or "codex".
pub system: Option<String>,
/// Git commit SHA where this skill was found (git hook path only).
pub commit_sha: Option<String>,
/// Git blob SHA-1 of the SKILL.md file.
pub blob_hash: Option<String>,
}
/// Compute the git blob SHA-1 hash of `content`.
/// Format: "blob {len}\0{data}"
fn git_blob_hash(content: &[u8]) -> String {
let size = content.len();
let header = format!("blob {}\0", size);
let mut hasher = sha1::Sha1::new();
hasher.update(header.as_bytes());
hasher.update(content);
hex::encode(hasher.finalize())
}
fn parse_frontmatter(frontmatter: Option<&str>) -> serde_json::Value {
frontmatter
.and_then(|fm| serde_json::from_str(fm).ok())
.or_else(|| frontmatter.and_then(|fm| serde_yaml::from_str(fm).ok()))
.unwrap_or_default()
}
/// Parse a SKILL.md file and extract metadata + content.
fn parse_skill_file(slug: &str, raw: &str) -> DiscoveredSkill {
let (frontmatter, content) = extract_frontmatter(raw);
let metadata = parse_frontmatter(frontmatter);
let name = metadata
.get("name")
.and_then(|v| v.as_str())
.map(String::from)
.unwrap_or_else(|| slug.replace('-', " ").replace('_', " "));
let description = metadata
.get("description")
.and_then(|v| v.as_str())
.map(String::from);
DiscoveredSkill {
slug: slug.to_string(),
name,
description,
content: content.trim().to_string(),
metadata,
relative_path: None,
system: None,
commit_sha: None,
blob_hash: None,
}
}
/// Split frontmatter (--- ... ---) from markdown content.
fn extract_frontmatter(raw: &str) -> (Option<&str>, &str) {
let trimmed = raw.trim_start();
if !trimmed.starts_with("---") {
return (None, trimmed);
}
if let Some(end) = trimmed[3..].find("---") {
let fm = &trimmed[3..end + 3];
let rest = trimmed[3 + end + 3..].trim_start();
(Some(fm), rest)
} else {
(None, trimmed)
}
}
/// Recursively scan supported skill roots for `SKILL.md` files (filesystem walk, non-bare repos).
/// The skill slug is `{short_repo_id}/{system}/{relative_skill_dir}`.
pub fn scan_repo_for_skills(
repo_path: &Path,
repo_id: Uuid,
) -> Result<Vec<DiscoveredSkill>, AppError> {
let repo_id_prefix = &repo_id.to_string()[..8];
let mut discovered = Vec::new();
for (root, system) in SKILL_ROOTS {
let root_path = repo_path.join(root);
if root_path.exists() {
scan_skill_root_fs(&root_path, repo_id_prefix, system, root, &mut discovered);
}
}
scan_root_skill_pack_fs(repo_path, repo_id_prefix, &mut discovered);
Ok(discovered)
}
fn scan_skill_root_fs(
root_path: &Path,
repo_id_prefix: &str,
system: &str,
root: &str,
discovered: &mut Vec<DiscoveredSkill>,
) {
let mut stack = vec![root_path.to_path_buf()];
while let Some(dir) = stack.pop() {
let entries = match std::fs::read_dir(&dir) {
Ok(e) => e,
Err(_) => continue,
};
for entry in entries.flatten() {
let path = entry.path();
if path.is_dir() {
stack.push(path);
continue;
}
if !is_skill_file_name(&path) {
continue;
}
let Some(parent) = path.parent() else {
continue;
};
let relative_skill_dir = parent
.strip_prefix(root_path)
.ok()
.and_then(path_to_slug)
.filter(|s| !s.is_empty());
let Some(relative_skill_dir) = relative_skill_dir else {
continue;
};
let slug = format!("{}/{}/{}", repo_id_prefix, system, relative_skill_dir);
if let Ok(raw) = std::fs::read(&path) {
let blob_hash = git_blob_hash(&raw);
let mut skill = parse_skill_file(&slug, &String::from_utf8_lossy(&raw));
skill.blob_hash = Some(blob_hash);
skill.system = Some(system.to_string());
skill.relative_path = Some(format!("{}/{}/SKILL.md", root, relative_skill_dir));
skill.metadata =
enrich_metadata(skill.metadata, system, skill.relative_path.as_deref());
discovered.push(skill);
}
}
}
}
fn scan_root_skill_pack_fs(
repo_path: &Path,
repo_id_prefix: &str,
discovered: &mut Vec<DiscoveredSkill>,
) {
let entries = match std::fs::read_dir(repo_path) {
Ok(entries) => entries,
Err(_) => return,
};
for entry in entries.flatten() {
let path = entry.path();
if !path.is_dir() {
continue;
}
let Some(dir_name) = path.file_name().and_then(|n| n.to_str()) else {
continue;
};
if dir_name == ".git" || dir_name == ".claude" || dir_name == ".codex" {
continue;
}
let skill_file = path.join("SKILL.md");
if !skill_file.exists() {
continue;
}
let relative_skill_dir = slugify_segment(dir_name);
if relative_skill_dir.is_empty() {
continue;
}
let slug = format!("{}/{}", repo_id_prefix, relative_skill_dir);
if let Ok(raw) = std::fs::read(&skill_file) {
let blob_hash = git_blob_hash(&raw);
let mut skill = parse_skill_file(&slug, &String::from_utf8_lossy(&raw));
skill.blob_hash = Some(blob_hash);
skill.system = Some(ROOT_SKILL_SYSTEM.to_string());
skill.relative_path = Some(format!("{}/SKILL.md", relative_skill_dir));
skill.metadata = enrich_metadata(
skill.metadata,
ROOT_SKILL_SYSTEM,
skill.relative_path.as_deref(),
);
discovered.push(skill);
}
}
}
/// Scan git tree objects for `SKILL.md` files (works for bare repos).
/// Traverses the HEAD commit tree using libgit2, reading blob content from objects.
pub fn scan_repo_tree_for_skills(
git_repo: &Repository,
repo_id: Uuid,
) -> Result<Vec<DiscoveredSkill>, AppError> {
let repo_id_prefix = &repo_id.to_string()[..8];
let head = git_repo
.head()
.map_err(|e| AppError::InternalServerError(format!("no HEAD: {e}")))?;
let tree = head
.peel_to_tree()
.map_err(|e| AppError::InternalServerError(format!("no tree: {e}")))?;
let mut discovered = Vec::new();
let mut stack: Vec<(git2::Tree<'_>, String)> = vec![(tree, String::new())];
while let Some((current_tree, prefix)) = stack.pop() {
for entry in current_tree.iter() {
let name = match entry.name() {
Some(n) => n,
None => continue,
};
let entry_path = if prefix.is_empty() {
name.to_string()
} else {
format!("{}/{}", prefix, name)
};
match entry.kind() {
Some(git2::ObjectType::Tree) => {
if should_descend_dir(name) {
if let Ok(subtree) =
entry.to_object(git_repo).and_then(|o| o.peel_to_tree())
{
stack.push((subtree, entry_path));
}
}
}
Some(git2::ObjectType::Blob) if name.eq_ignore_ascii_case("SKILL.md") => {
let Some((system, relative_skill_dir, legacy_slug)) =
skill_location_from_path(&entry_path)
else {
continue;
};
let slug = if legacy_slug {
format!("{}/{}", repo_id_prefix, relative_skill_dir)
} else {
format!("{}/{}/{}", repo_id_prefix, system, relative_skill_dir)
};
if let Ok(blob) = entry.to_object(git_repo).and_then(|o| o.peel_to_blob()) {
let raw = blob.content();
let blob_hash = git_blob_hash(raw);
let mut skill = parse_skill_file(&slug, &String::from_utf8_lossy(raw));
skill.blob_hash = Some(blob_hash);
skill.system = Some(system.to_string());
skill.relative_path = Some(entry_path.clone());
skill.metadata =
enrich_metadata(skill.metadata, system, skill.relative_path.as_deref());
discovered.push(skill);
}
}
_ => {}
}
}
}
Ok(discovered)
}
fn is_skill_file_name(path: &Path) -> bool {
path.file_name()
.and_then(|n| n.to_str())
.is_some_and(|name| name.eq_ignore_ascii_case("SKILL.md"))
}
fn path_to_slug(path: &Path) -> Option<String> {
let parts: Vec<String> = path
.components()
.filter_map(|c| c.as_os_str().to_str())
.map(slugify_segment)
.filter(|s| !s.is_empty())
.collect();
(!parts.is_empty()).then(|| parts.join("/"))
}
fn slugify_segment(input: &str) -> String {
let mut out = String::with_capacity(input.len());
let mut last_dash = false;
for ch in input.chars() {
let ch = ch.to_ascii_lowercase();
if ch.is_ascii_alphanumeric() || ch == '_' || ch == '-' {
out.push(ch);
last_dash = false;
} else if !last_dash {
out.push('-');
last_dash = true;
}
}
out.trim_matches('-').to_string()
}
fn skill_location_from_path(path: &str) -> Option<(&'static str, String, bool)> {
let normalized = path.replace('\\', "/");
for (root, system) in SKILL_ROOTS {
let prefix = format!("{}/", root);
let suffix = "/SKILL.md";
if normalized.starts_with(&prefix) && normalized.ends_with(suffix) {
let relative = &normalized[prefix.len()..normalized.len() - suffix.len()];
let slug = relative
.split('/')
.map(slugify_segment)
.filter(|s| !s.is_empty())
.collect::<Vec<_>>()
.join("/");
if !slug.is_empty() {
return Some((*system, slug, false));
}
}
}
let suffix = "/SKILL.md";
if normalized.ends_with(suffix) && !normalized.starts_with('.') {
let relative = &normalized[..normalized.len() - suffix.len()];
if !relative.contains('/') {
let slug = slugify_segment(relative);
if !slug.is_empty() {
return Some((ROOT_SKILL_SYSTEM, slug, true));
}
}
}
None
}
fn enrich_metadata(
mut metadata: serde_json::Value,
system: &str,
relative_path: Option<&str>,
) -> serde_json::Value {
if !metadata.is_object() {
metadata = serde_json::json!({});
}
if let Some(obj) = metadata.as_object_mut() {
obj.entry("system")
.or_insert_with(|| serde_json::Value::String(system.to_string()));
if let Some(relative_path) = relative_path {
obj.entry("path")
.or_insert_with(|| serde_json::Value::String(relative_path.to_string()));
}
}
metadata
}
#[cfg(test)]
mod tests {
use super::{scan_repo_for_skills, scan_repo_tree_for_skills};
use git2::{Repository, Signature};
use std::fs;
use tempfile::tempdir;
use uuid::Uuid;
#[test]
fn tree_scan_discovers_skills_under_hidden_dirs() {
let temp = tempdir().expect("tempdir");
let repo = Repository::init(temp.path()).expect("init repo");
let skill_dir = temp
.path()
.join(".claude")
.join("skills")
.join("demo-skill");
fs::create_dir_all(&skill_dir).expect("create skill dir");
fs::write(
skill_dir.join("SKILL.md"),
"---\nname: Demo Skill\ndescription: test\n---\ncontent",
)
.expect("write skill");
let mut index = repo.index().expect("index");
index
.add_path(std::path::Path::new(".claude/skills/demo-skill/SKILL.md"))
.expect("add skill");
index.write().expect("write index");
let tree_id = index.write_tree().expect("write tree");
let tree = repo.find_tree(tree_id).expect("find tree");
let sig = Signature::now("tester", "tester@example.com").expect("signature");
repo.commit(Some("HEAD"), &sig, &sig, "add skill", &tree, &[])
.expect("commit");
let discovered = scan_repo_tree_for_skills(&repo, Uuid::nil()).expect("scan tree");
assert_eq!(discovered.len(), 1);
assert_eq!(discovered[0].name, "Demo Skill");
assert_eq!(discovered[0].slug, "00000000/claude/demo-skill");
assert_eq!(
discovered[0].metadata["system"],
serde_json::Value::String("claude".into())
);
}
#[test]
fn tree_scan_keeps_claude_and_codex_skills_separate() {
let temp = tempdir().expect("tempdir");
let repo = Repository::init(temp.path()).expect("init repo");
for root in [".claude", ".codex"] {
let skill_dir = temp.path().join(root).join("skills").join("review");
fs::create_dir_all(&skill_dir).expect("create skill dir");
fs::write(
skill_dir.join("SKILL.md"),
format!("---\nname: {} Review\n---\ncontent", root),
)
.expect("write skill");
}
let mut index = repo.index().expect("index");
index
.add_path(std::path::Path::new(".claude/skills/review/SKILL.md"))
.expect("add claude skill");
index
.add_path(std::path::Path::new(".codex/skills/review/SKILL.md"))
.expect("add codex skill");
index.write().expect("write index");
let tree_id = index.write_tree().expect("write tree");
let tree = repo.find_tree(tree_id).expect("find tree");
let sig = Signature::now("tester", "tester@example.com").expect("signature");
repo.commit(Some("HEAD"), &sig, &sig, "add skills", &tree, &[])
.expect("commit");
let mut slugs = scan_repo_tree_for_skills(&repo, Uuid::nil())
.expect("scan tree")
.into_iter()
.map(|s| s.slug)
.collect::<Vec<_>>();
slugs.sort();
assert_eq!(
slugs,
vec!["00000000/claude/review", "00000000/codex/review"]
);
}
#[test]
fn scans_root_level_skill_pack_layout() {
let temp = tempdir().expect("tempdir");
let repo = Repository::init(temp.path()).expect("init repo");
let skill_dir = temp.path().join("code-review");
fs::create_dir_all(&skill_dir).expect("create skill dir");
fs::write(skill_dir.join("SKILL.md"), "# Code Review\n\ncontent").expect("write skill");
let mut index = repo.index().expect("index");
index
.add_path(std::path::Path::new("code-review/SKILL.md"))
.expect("add skill");
index.write().expect("write index");
let tree_id = index.write_tree().expect("write tree");
let tree = repo.find_tree(tree_id).expect("find tree");
let sig = Signature::now("tester", "tester@example.com").expect("signature");
repo.commit(Some("HEAD"), &sig, &sig, "add skill", &tree, &[])
.expect("commit");
let fs_discovered = scan_repo_for_skills(temp.path(), Uuid::nil()).expect("scan fs");
let tree_discovered = scan_repo_tree_for_skills(&repo, Uuid::nil()).expect("scan tree");
assert_eq!(fs_discovered.len(), 1);
assert_eq!(tree_discovered.len(), 1);
assert_eq!(fs_discovered[0].slug, "00000000/code-review");
assert_eq!(tree_discovered[0].slug, "00000000/code-review");
assert_eq!(
tree_discovered[0].metadata["system"],
serde_json::Value::String("root".into())
);
}
}
/// Scan a git2::Repository for skills and upsert them into the database.
/// Uses filesystem walk for normal repos, git tree traversal for bare repos.
pub async fn scan_and_sync_skills(
db: &db::database::AppDatabase,
project_uuid: Uuid,
repo: &RepoModel,
) -> Result<ScanSyncResult, AppError> {
// Open with git2 to get the actual workdir
let git_repo = match Repository::open(&repo.storage_path) {
Ok(r) => r,
Err(e) => {
tracing::warn!("failed to open git repo {}: {:?}", repo.storage_path, e);
return Ok(ScanSyncResult {
discovered: 0,
created: 0,
updated: 0,
removed: 0,
});
}
};
let commit_sha = git_repo
.head()
.ok()
.and_then(|h| h.target())
.map(|oid| oid.to_string());
// For bare repos (no workdir), scan git tree objects directly
let mut discovered = if git_repo.is_bare() || git_repo.workdir().is_none() {
match scan_repo_tree_for_skills(&git_repo, repo.id) {
Ok(skills) => skills,
Err(e) => {
tracing::warn!("tree scan failed for repo {}: {:?}", repo.storage_path, e);
vec![]
}
}
} else {
let workdir = git_repo.workdir().unwrap();
scan_repo_for_skills(workdir, repo.id)?
};
// Fill in commit_sha for discovered skills
for skill in &mut discovered {
skill.commit_sha = commit_sha.clone();
}
sync_discovered_skills(db, project_uuid, repo.id, discovered).await
}
/// Sync discovered skills by stable slug.
async fn sync_discovered_skills(
db: &db::database::AppDatabase,
project_uuid: Uuid,
repo_id: Uuid,
discovered: Vec<DiscoveredSkill>,
) -> Result<ScanSyncResult, AppError> {
if discovered.is_empty() {
return Ok(ScanSyncResult {
discovered: 0,
created: 0,
updated: 0,
removed: 0,
});
}
let now = Utc::now();
let mut created = 0i64;
let mut updated = 0i64;
// Deduplicate by slug. The slug includes repo prefix + skill system + relative skill path.
let mut deduped: std::collections::HashMap<String, DiscoveredSkill> =
std::collections::HashMap::new();
for skill in discovered {
match deduped.get(&skill.slug) {
Some(existing) => {
// Keep the one with the later commit_sha
if skill.commit_sha.as_ref().unwrap_or(&String::new())
> existing.commit_sha.as_ref().unwrap_or(&String::new())
{
deduped.insert(skill.slug.clone(), skill);
}
}
None => {
deduped.insert(skill.slug.clone(), skill);
}
}
}
// Query existing skills for this repo
let existing: Vec<_> = SkillEntity::find()
.filter(C::ProjectUuid.eq(project_uuid))
.filter(C::Source.eq("repo"))
.filter(C::RepoId.eq(repo_id))
.all(db)
.await?;
let existing_by_slug: std::collections::HashMap<_, _> =
existing.into_iter().map(|s| (s.slug.clone(), s)).collect();
let mut seen_keys = std::collections::HashSet::new();
let discovered_count = deduped.len() as i64;
for (key, skill) in deduped {
seen_keys.insert(key.clone());
let json_meta = serde_json::to_value(&skill.metadata).unwrap_or_default();
if let Some(existing_skill) = existing_by_slug.get(&key) {
if existing_skill.content != skill.content
|| existing_skill.metadata != json_meta
|| existing_skill.commit_sha != skill.commit_sha
|| existing_skill.blob_hash != skill.blob_hash
|| existing_skill.name != skill.name
|| existing_skill.description != skill.description
{
let mut active: SkillActiveModel = existing_skill.clone().into();
active.name = Set(skill.name);
active.description = Set(skill.description);
active.content = Set(skill.content);
active.metadata = Set(json_meta);
active.commit_sha = Set(skill.commit_sha);
active.blob_hash = Set(skill.blob_hash);
active.updated_at = Set(now);
active.update(db).await?;
updated += 1;
}
} else {
let active = SkillActiveModel {
id: Set(0),
project_uuid: Set(project_uuid),
slug: Set(skill.slug),
name: Set(skill.name),
description: Set(skill.description),
source: Set("repo".to_string()),
repo_id: Set(Some(repo_id)),
commit_sha: Set(skill.commit_sha),
blob_hash: Set(skill.blob_hash),
content: Set(skill.content),
metadata: Set(json_meta),
enabled: Set(true),
created_by: Set(None),
created_at: Set(now),
updated_at: Set(now),
};
active.insert(db).await?;
created += 1;
}
}
// Remove skills that no longer exist in the repo
let mut removed = 0i64;
for (key, old_skill) in existing_by_slug {
if !seen_keys.contains(&key) {
SkillEntity::delete_by_id(old_skill.id).exec(db).await?;
removed += 1;
}
}
Ok(ScanSyncResult {
discovered: discovered_count,
created,
updated,
removed,
})
}
/// Result of a scan + sync operation.
#[derive(Debug)]
pub struct ScanSyncResult {
pub discovered: i64,
pub created: i64,
pub updated: i64,
pub removed: i64,
}