osborn 0.9.22 → 0.9.24

package/dist/config.js

@@ -649,11 +649,31 @@ export async function listAllClaudeSessions(limit = 100) {
             ]);
             if (preview.messageCount < 2)
                 continue;
+            // SLUG-FIRST: prefer the slug-derived path over the content cwd.
+            //
+            // Claude Code's `--resume <id>` looks up sessions by file LOCATION
+            // (slug folder under ~/.claude/projects/), not by the cwd field inside
+            // the JSONL. The JSONL content cwd is immutable historical metadata —
+            // it's where the session was ORIGINALLY recorded (e.g. sprite path
+            // /home/sprite/workspace or /workspaces/codespaces-blank). On migration
+            // (sprite → fly), files were finalized into a new slug but their
+            // content still records the original cwd.
+            //
+            // The dashboard forwards this field as `workingDirectory` to the
+            // agent, which uses it as Claude Code's spawn cwd. If it doesn't
+            // match the slug, the spawn lands in the wrong folder and resume
+            // errors with "No conversation found". The slug is the source of
+            // truth for where the file lives on disk; use it.
+            //
+            // Content cwd is kept as a last-resort fallback only when slug
+            // reversal fails (slugToPath validates with existsSync and returns
+            // '' on ambiguous encodings like dir names containing literal '-').
+            const slugPath = slugToPath(c.slug);
             sessions.push({
                 sessionId: c.sessionId,
                 projectSlug: c.slug,
-                projectPath: cwd || slugToPath(c.slug),
-                cwd: cwd || slugToPath(c.slug),
+                projectPath: slugPath || cwd,
+                cwd: slugPath || cwd,
                 timestamp: c.mtime,
                 lastMessage: preview.lastMessage,
                 messageCount: preview.messageCount,

package/dist/index.js

@@ -323,8 +323,13 @@ function startApiServer(workingDir, port) {
             console.log('[events] SSE client connected');
             return;
         }
-        // GET /sessions/export — stream a gzipped tar of ~/.claude/projects/ to the client
-        // Optional ?workDir= query param: if present, export only that project's slug folder.
+        // GET /sessions/export — stream a gzipped tar of ~/.claude/projects/ AND
+        // ~/.claude/skills/ to the client. Both directories ship in one archive so
+        // a sync covers conversations (projects/) and learned skills together —
+        // e.g. PostCompact-written `decisions/SKILL.md` and `learned-behaviors/SKILL.md`
+        // travel with the user's session data.
+        // Optional ?workDir= query param accepted for backwards compat but ignored
+        // (full export is always returned).
         if (req.method === 'GET' && url.pathname === '/sessions/export') {
             if (syncToken) {
                 const authHeader = req.headers['authorization'] ?? '';
@@ -336,17 +341,24 @@ function startApiServer(workingDir, port) {
             }
             const claudeDir = join(homedir(), '.claude');
             const projectsDir = join(claudeDir, 'projects');
+            const skillsDir = join(claudeDir, 'skills');
             const workDir = url.searchParams.get('workDir');
-            if (!existsSync(projectsDir)) {
+            void workDir;
+            // Collect which top-level dirs exist — tar fails if we list one that doesn't.
+            const topLevel = [];
+            if (existsSync(projectsDir))
+                topLevel.push('projects');
+            if (existsSync(skillsDir))
+                topLevel.push('skills');
+            if (topLevel.length === 0) {
                 res.writeHead(404, { 'Content-Type': 'application/json' });
-                res.end(JSON.stringify({ error: 'No sessions found' }));
+                res.end(JSON.stringify({ error: 'No sessions or skills found' }));
                 return;
             }
-            const tarArgs = ['-czf', '-', '-C', claudeDir, 'projects'];
-            void workDir; // workDir param accepted but full projects/ export is returned
+            const tarArgs = ['-czf', '-', '-C', claudeDir, ...topLevel];
             res.writeHead(200, {
                 'Content-Type': 'application/gzip',
-                'Content-Disposition': 'attachment; filename="claude-sessions.tar.gz"',
+                'Content-Disposition': 'attachment; filename="claude-export.tar.gz"',
                 'Access-Control-Allow-Origin': '*',
             });
             // Stream tar output directly to response
@@ -358,23 +370,39 @@ function startApiServer(workingDir, port) {
             return;
         }
         // GET /sessions/manifest — return mtime+size for all .jsonl files per slug (public, no auth)
-        // Helper: merge an extracted tar directory into ~/.claude/projects/ with all 4 fixes:
+        // Helper: merge an extracted tar directory into ~/.claude/{projects,skills}/.
+        //
+        // Behavior:
         //   1. Skip macOS AppleDouble entries (`._*`) that bsdtar emits
-        //   2. Apply slug remap when targetWorkDir is supplied (chunked path missed this)
-        //   3. Rewrite embedded `cwd` field inside .jsonl entries during remap so
-        //      Claude Code can resume the conversation in the destination workspace
-        //   4. Merge into existing dest dirs instead of failing on rename collision
-        const mergeExtractedIntoProjects = async (sourceDir, targetWorkDir) => {
+        //   2. Project slugs: apply slug remap when targetWorkDir is supplied so
+        //      laptop/codespaces sessions land at the destination's slug.
+        //   3. Skills: copy each <skillName>/ directory verbatim (no slug remap —
+        //      skill identity is the directory name, same across all environments).
+        //      PostCompact-learned skills like 'decisions' and 'learned-behaviors'
+        //      from a sprite travel through to the destination this way.
+        //   4. Per-file mtime-newer-wins resolves collisions in either dir (preserves
+        //      whichever side has the more-recent version).
+        //   5. Byte-exact copy — no content mutation in either projects/ or skills/.
+        const mergeExtractedClaudeDir = async (sourceDir, targetWorkDir) => {
             const claudeDir = join(homedir(), '.claude');
             const projectsDir = join(claudeDir, 'projects');
+            const skillsDir = join(claudeDir, 'skills');
             mkdirSync(projectsDir, { recursive: true });
-            // The archive sometimes wraps content in a 'projects' subdir, sometimes not.
+            mkdirSync(skillsDir, { recursive: true });
+            // The archive may wrap content in a 'projects' / 'skills' subdir, or be
+            // a flat dir of slugs. Detect both.
             const extractedProjects = join(sourceDir, 'projects');
-            const effectiveSource = existsSync(extractedProjects) ? extractedProjects : sourceDir;
-            // Filter out AppleDouble (`._*`) entries that macOS bsdtar emits for
-            // resource forks. These crash later steps if they collide with real dirs.
-            const sourceSlugs = readdirSync(effectiveSource)
-                .filter(s => !s.startsWith('._') && !s.startsWith('.DS_Store'));
+            const extractedSkills = join(sourceDir, 'skills');
+            const hasProjectsWrapper = existsSync(extractedProjects);
+            const hasSkillsWrapper = existsSync(extractedSkills);
+            // If neither wrapper exists, treat the source dir as a flat slug dir
+            // (back-compat with older client tar layouts).
+            const effectiveSource = (hasProjectsWrapper || hasSkillsWrapper) ? null : sourceDir;
+            // ─── PROJECTS extraction ─────────────────────────────────────────
+            const projectsSource = hasProjectsWrapper ? extractedProjects : effectiveSource;
+            const sourceSlugs = projectsSource
+                ? readdirSync(projectsSource).filter(s => !s.startsWith('._') && !s.startsWith('.DS_Store'))
+                : [];
             // Build remap table: source-slug → target-slug.
             // Only remaps slugs that differ from the target (no-op if already correct).
             const remapped = {};
@@ -396,7 +424,7 @@ function startApiServer(workingDir, port) {
                 const effectiveSlug = remapped[sourceSlug] ?? sourceSlug;
                 const destSlug = join(projectsDir, effectiveSlug);
                 mkdirSync(destSlug, { recursive: true });
-                const sourceSlugPath = join(effectiveSource, sourceSlug);
+                const sourceSlugPath = join(projectsSource, sourceSlug);
                 // NO content mutation. Earlier versions rewrote the embedded `"cwd":"..."`
                 // field inside JSONL entries to match the destination workspace. That was
                 // wrong on two counts:
@@ -460,27 +488,71 @@ function startApiServer(workingDir, port) {
                     catch { /* ignore */ }
                 }
             }
-            return { filesWritten, remapped };
+            // ─── SKILLS extraction ───────────────────────────────────────────
+            // Skill identity is the directory name (`decisions`, `learned-behaviors`,
+            // etc.) so there's no slug remap — each <skillName>/ dir copies into
+            // ~/.claude/skills/<skillName>/. Mtime-newer-wins handles collisions:
+            //   - Default skills seeded by Docker entrypoint have the boot mtime
+            //   - Learned skills from source have whatever mtime they had on origin
+            //   - Newer side wins per file
+            let skillsWritten = 0;
+            if (hasSkillsWrapper) {
+                const sourceSkillNames = readdirSync(extractedSkills)
+                    .filter(s => !s.startsWith('._') && !s.startsWith('.DS_Store'));
+                for (const skillName of sourceSkillNames) {
+                    const srcSkillPath = join(extractedSkills, skillName);
+                    const dstSkillPath = join(skillsDir, skillName);
+                    mkdirSync(dstSkillPath, { recursive: true });
+                    // Reuse the same mtime-aware walkAndCopy from the projects loop —
+                    // it's still in scope from the last iteration. If sourceSlugs was
+                    // empty, define it inline here. (Define standalone to be safe.)
+                    const walkSkill = (src, dst) => {
+                        const entries = readdirSync(src, { withFileTypes: true });
+                        for (const e of entries) {
+                            if (e.name.startsWith('._') || e.name === '.DS_Store')
+                                continue;
+                            const sp = join(src, e.name);
+                            const dp = join(dst, e.name);
+                            if (e.isDirectory()) {
+                                mkdirSync(dp, { recursive: true });
+                                walkSkill(sp, dp);
+                            }
+                            else if (e.isFile()) {
+                                let shouldWrite = true;
+                                try {
+                                    const dstStat = statSync(dp);
+                                    const srcStat = statSync(sp);
+                                    if (dstStat.mtimeMs >= srcStat.mtimeMs)
+                                        shouldWrite = false;
+                                }
+                                catch { }
+                                if (!shouldWrite)
+                                    continue;
+                                cpSync(sp, dp, { force: true });
+                                skillsWritten++;
+                            }
+                        }
+                    };
+                    walkSkill(srcSkillPath, dstSkillPath);
+                }
+            }
+            return { filesWritten, remapped, skillsWritten };
         };
         if (req.method === 'GET' && url.pathname === '/sessions/manifest') {
-            // Walks the FULL tree per slug — including sub-agent transcripts
-            // (<slug>/<sessionId>/subagents/*.jsonl), tool-results (<slug>/<sessionId>/tool-results/*),
-            // osb workspace files (<slug>/osb/<sessionId>/*), and file-history. Files are
-            // keyed by their path RELATIVE to the slug dir so the client can preserve
-            // structure when computing diffs. mtime is in ms epoch so a simple `>`
-            // comparison is the "newer wins" merge rule.
-            //
-            // Previous version only listed top-level *.jsonl and missed ~270/290 files
-            // on a typical session — sub-agent transcripts invisible → resume failed
-            // silently because Claude couldn't find the referenced agent_id transcripts.
-            const claudeDir = join(homedir(), '.claude', 'projects');
-            const slugMap = {};
-            const walkSlug = (slugDir) => {
+            // Walks BOTH ~/.claude/projects/ AND ~/.claude/skills/ — full tree per
+            // top-level directory.
+            //   - projects/<slug>/<...> → session JSONLs, sub-agent transcripts, tool-results, osb/
+            //   - skills/<skillName>/<...> → SKILL.md + any subfiles
+            // Files keyed by path RELATIVE to the slug/skill-name dir so the client
+            // can preserve structure when computing diffs. mtime in ms epoch.
+            const projectsRoot = join(homedir(), '.claude', 'projects');
+            const skillsRoot = join(homedir(), '.claude', 'skills');
+            const walkDir = (dir) => {
                 const files = {};
-                const walk = (dir, relPrefix) => {
+                const walk = (curr, relPrefix) => {
                     let entries;
                     try {
-                        entries = readdirSync(dir, { withFileTypes: true });
+                        entries = readdirSync(curr, { withFileTypes: true });
                     }
                     catch {
                         return;
@@ -488,11 +560,10 @@ function startApiServer(workingDir, port) {
                     for (const e of entries) {
                         if (e.name.startsWith('._') || e.name === '.DS_Store')
                             continue;
-                        const sub = join(dir, e.name);
+                        const sub = join(curr, e.name);
                         const rel = relPrefix ? `${relPrefix}/${e.name}` : e.name;
-                        if (e.isDirectory()) {
+                        if (e.isDirectory())
                             walk(sub, rel);
-                        }
                         else if (e.isFile()) {
                             try {
                                 const st = statSync(sub);
@@ -502,22 +573,29 @@ function startApiServer(workingDir, port) {
                         }
                     }
                 };
-                walk(slugDir, '');
+                walk(dir, '');
                 return files;
             };
+            const slugMap = {};
             try {
-                const slugs = readdirSync(claudeDir, { withFileTypes: true })
+                for (const slug of readdirSync(projectsRoot, { withFileTypes: true })
                     .filter(d => d.isDirectory() && !d.name.startsWith('._'))
-                    .map(d => d.name);
-                for (const slug of slugs) {
-                    slugMap[slug] = { files: walkSlug(join(claudeDir, slug)) };
+                    .map(d => d.name)) {
+                    slugMap[slug] = { files: walkDir(join(projectsRoot, slug)) };
                 }
             }
-            catch {
-                // projects dir doesn't exist yet — return empty
+            catch { /* projects dir doesn't exist yet — leave empty */ }
+            const skillsMap = {};
+            try {
+                for (const name of readdirSync(skillsRoot, { withFileTypes: true })
+                    .filter(d => d.isDirectory() && !d.name.startsWith('._'))
+                    .map(d => d.name)) {
+                    skillsMap[name] = { files: walkDir(join(skillsRoot, name)) };
+                }
             }
+            catch { /* skills dir doesn't exist yet — leave empty */ }
             res.writeHead(200, { 'Content-Type': 'application/json' });
-            res.end(JSON.stringify({ slugs: slugMap }));
+            res.end(JSON.stringify({ slugs: slugMap, skills: skillsMap }));
             return;
         }
         // POST /sessions/import — accept a gzipped tar and extract into ~/.claude/projects/
@@ -574,9 +652,9 @@ function startApiServer(workingDir, port) {
                         res.end(JSON.stringify({ error: 'tar extraction failed', code }));
                         return;
                     }
-                    const { filesWritten, remapped } = await mergeExtractedIntoProjects(tmpDir, targetWorkDir ?? undefined);
+                    const { filesWritten, remapped, skillsWritten } = await mergeExtractedClaudeDir(tmpDir, targetWorkDir ?? undefined);
                     res.writeHead(200, { 'Content-Type': 'application/json' });
-                    res.end(JSON.stringify({ ok: true, filesWritten, remapped }));
+                    res.end(JSON.stringify({ ok: true, filesWritten, remapped, skillsWritten }));
                 }
                 catch (err) {
                     console.error('[import] merge error:', err);
@@ -690,9 +768,9 @@ function startApiServer(workingDir, port) {
                             res.end(JSON.stringify({ error: 'tar extraction failed', code }));
                             return;
                         }
-                        const { filesWritten, remapped } = await mergeExtractedIntoProjects(tmpExtractDir, targetWorkDir);
+                        const { filesWritten, remapped, skillsWritten } = await mergeExtractedClaudeDir(tmpExtractDir, targetWorkDir);
                         res.writeHead(200, { 'Content-Type': 'application/json' });
-                        res.end(JSON.stringify({ ok: true, filesWritten, remapped }));
+                        res.end(JSON.stringify({ ok: true, filesWritten, remapped, skillsWritten }));
                     }
                     catch (err) {
                         console.error('[import-finalize] merge error:', err);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "osborn",
-  "version": "0.9.22",
+  "version": "0.9.24",
   "description": "Voice AI coding assistant - local agent that connects to Osborn frontend",
   "type": "module",
   "bin": {