Hermes stores reasoning as a top-level message field (m.reasoning)
instead of in content arrays like Claude. This patch makes the
thinking/reasoning card also check for m.reasoning, so users can
see the model's reasoning process in the WebUI.
Tool cards disappeared on page refresh because assistant messages with
only tool_use content (no text) were filtered out of the visible
messages list. Since tool cards anchor to DOM rows via data-msg-idx,
removing the anchor row meant cards had nothing to attach to.
Fix: keep assistant messages in the render list if they contain
tool_use blocks, even when they have no text content. The row renders
with the role label but empty body, providing an anchor point for the
tool card insertion pass.
Fixes#140
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: tool call cards persist correctly after page reload
Root cause: the insertion logic looked for the NEXT assistant row to
insert BEFORE, but when the triggering assistant message contained only
tool_use blocks (no text), it was filtered from the DOM by msgContent()
and no anchor row existed. Cards were silently dropped.
Fix: find the row AT the assistant_msg_idx first (exact match), fall
back to the nearest preceding visible row, then insert AFTER it (not
before the next). This handles both text+tool and tool-only assistant
messages correctly.
Closes#140
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: tool card ordering — role-filter fallback anchor, preserve order for same-anchor groups
Two bugs in the initial PR:
- Fallback 'nearest row' had no role filter, could anchor to a user message
row causing cards to appear after a user bubble. Fixed: check role==='assistant'
in the fallback loop.
- Back-to-back tool-card groups sharing a filtered anchor were inserted in
reverse order because each group re-read anchorRow.nextSibling from the live
DOM. Fixed: track the last inserted node per anchor via anchorInsertAfter Map,
so each subsequent group appends after the previous one.
Also restores the 'Collect card elements before they get moved to DOM' comment
explaining why Array.from() is used on frag before DOM insertion.
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Move `import html` to module top (was inside function body)
- Fix IndexError crash in /login when bot_name is empty string;
use `or 'Hermes'` fallback instead of .get() default which
doesn't guard against stored empty string
- Add server-side sanitization in POST /api/settings: strip + default
empty/whitespace bot_name to 'Hermes' before persisting
- Escape _bn initial char in ui.js innerHTML (esc() consistency)
- Add maxlength=64 to #settingsBotName input field
- Add tests/test_sprint27.py: 9 tests covering API round-trip,
empty/whitespace defaults, login page rendering, and XSS escaping
Shows a blue banner when the webui or hermes-agent git repos are behind
their upstream branches. One-click 'Update Now' button does stash, pull
--ff-only, stash pop, then reloads the page.
Backend (api/updates.py):
- _check_repo(): git fetch + rev-list count with 15s timeout
- check_for_updates(): 30-min server-side cache, thread-safe, skips
Docker (no .git dir)
- apply_update(): stash (if dirty), pull --ff-only, pop, invalidate cache
Routes:
- GET /api/updates/check -- returns cached {webui, agent} with behind count
- POST /api/updates/apply -- {target: 'webui'|'agent'}
Frontend:
- Blue banner (matches reconnect-banner pattern) with 'Later' / 'Update Now'
- Non-blocking boot check via fire-and-forget .then(), once per tab session
- sessionStorage guards prevent re-checking and re-showing after dismiss
Settings:
- 'Check for updates' checkbox (default: on) -- when off, no git operations
- Removed 'Default Workspace' dropdown to keep settings panel compact
Performance:
- Server cache: git fetch at most 2x/hour regardless of client count
- sessionStorage: one check per browser tab session
- _check_in_progress flag prevents concurrent fetch storms
- Fire-and-forget: does NOT block the boot sequence
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The context indicator in the composer footer now shows real data from
the agent's context compressor instead of hardcoded estimates:
- last_prompt_tokens / context_length (e.g. '12.4k / 200k (6%)')
- Bar color: blue <50%, yellow 50-75%, red >75%
- Hover tooltip shows exact numbers + compression threshold
- Cost appended when available
Backend: streaming.py now reads context_length, threshold_tokens, and
last_prompt_tokens from agent.context_compressor after run_conversation()
and includes them in the usage dict sent with the 'done' SSE event.
This matches the CLI's context window display (the bar that shows
current context vs total window).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Agent review: hardcoded 128000 is wrong for Claude (200k), Gemini (1M),
and smaller models (8k-32k). Added a lookup table keyed by model name
substring covering major families with 128k fallback. TODO comment
for fetching exact values from server.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Shows a compact bar + label in the composer footer after the first
response, displaying input/output token counts, context window fill
percentage, and estimated cost. Bar turns yellow >50% and red >75%.
Updates on every response completion via the existing usage data from
the done SSE event. Hidden until first response (no usage data yet).
Inspired by PR #75 (@MartinNielsenDev).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Token usage display:
- Add 'show_token_usage' boolean to settings (default: false, off by default)
- Settings panel: checkbox 'Show token usage after responses'
- /usage slash command: instant toggle with toast feedback, persists to
server, updates checkbox if settings panel is open, re-renders messages
- Boot: load show_token_usage alongside send_key on startup
- ui.js: gate usage badge on window._showTokenUsage flag
Timestamps:
- streaming.py: stamp 'timestamp' on every message that lacks one at
conversation completion; old messages (no timestamp field) now get a
wall-clock time the first time they're touched by a new turn
- messages.js: stamp _ts on the last assistant message at done-event time
so the time shows immediately on the current turn before next reload
- Timestamps already render in the UI (Sprint 14): faint time on each
role header line, full opacity on hover, full date in title tooltip
Store expanded directory paths in localStorage keyed by workspace path
(key: 'hermes-webui-expanded:{workspacePath}'). On root load (loadDir('.')),
restore the saved set for the current workspace and pre-fetch dir contents
for any restored expanded directories so the tree renders fully on first
paint without requiring a second click to expand.
Saves on every expand/collapse toggle. Switching workspaces automatically
picks up that workspace's own saved state. Per-workspace (not per-session)
so the same tree state is shared across sessions using the same workspace,
which is the natural expectation.
- routes.py: reject glob wildcards (* ? [ ]) in skill name param to
prevent rglob wildcard injection when serving linked files
- panels.js: replace inline onclick+esc() with data-* attributes and
addEventListener for skill tag removal and linked-file clicks;
esc() is HTML-safe but not JS-safe -- apostrophes in names caused
JS syntax errors and _cronSelectedSkills array corruption
- ui.js: fix _fmtTokens(null/undefined) returning 'null'/'undefined'
by guarding with (!n||n<0) -> '0'; add data-role attribute to msg-row
elements so usage badge correctly targets the last assistant row
instead of the last row regardless of speaker
- tests: rename test_sprint24.py -> test_sprint23.py (wrong sprint #);
add 3 new tests: path traversal rejection, wildcard name rejection,
cron create with skills; strengthen existing tests to assert field
presence explicitly (was using .get(field, 0)==0 which never caught
a missing field)
Track A: Token/cost display
- Read agent usage attrs (session_prompt_tokens, session_completion_tokens,
session_estimated_cost_usd) after run_conversation in streaming.py
- Add input_tokens, output_tokens, estimated_cost fields to Session model
- Include usage in done SSE event payload
- Store usage on S.lastUsage in messages.js done handler
- Render usage badge below last assistant message (input/output/cost)
Track B: Subagent delegation cards
- Add subagent_progress to toolIcon map with shuffle emoji
- Special-case subagent_progress in buildToolCard: "Subagent" label,
strip double emoji from preview, add tool-card-subagent CSS class
- Indented border-left styling for subagent cards
- Clean delegate_task display name
Track C: Skill picker in cron create form
- Add skill search input + tag chips to cron create form HTML
- Skill picker JS in panels.js: search/filter, click-to-add tags,
remove tag chips, pre-fetch skill list on form open
- submitCronCreate sends skills array in POST body
- Skill picker dropdown + tag CSS
Track D: Skill linked files viewer
- Add file query param to /api/skills/content endpoint
- Serve linked files from skill directory with path traversal protection
- Ensure linked_files key always present in skill content response
- Render linked files section below SKILL.md content in preview panel
- openSkillFile function for viewing individual linked files
Track E: Bug fixes and code quality
- Expand Session.__init__ and compact() to readable multi-line format
- Remove inline import json as _j2 inside loop in streaming.py
- Fix tool_calls: capture args from assistant messages, skip unresolved names
- Store args snapshot in persisted tool_calls for reload display
6 new tests. Total: 421 (409 passing).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Root cause: three interacting bugs caused the model picker to show the wrong
model or flicker after a profile switch.
Bug 1 — syncTopbar() fought switchToProfile().
After switchToProfile() set the picker to the profile's model, syncTopbar()
was called (via renderSessionList -> loadSession, then explicitly at the end)
and overwrote it with S.session.model -- the old session's model.
Fix: added S._pendingProfileModel flag. switchToProfile() sets it;
syncTopbar() checks it first, applies the override, then clears it.
S.session.model is also updated to the resolved value so subsequent
syncTopbar() calls are consistent.
Bug 2 — Raw option injected at top of list for mismatched model IDs.
Profile configs store model IDs like 'claude-sonnet-4-6' (hermes-agent
format: hyphens, no namespace prefix) but the dropdown has
'anthropic/claude-sonnet-4.6' (OpenRouter format: dots, with prefix).
The old code did sel.value = id, found no match, then injected a new
<option> at the top of the list -- creating a lowercase duplicate that
didn't match any real provider group entry.
Fix: _findModelInDropdown() normalises both sides (strip prefix, hyphens->dots,
lowercase) and finds the best matching existing option. No new options are ever
injected for profile switching.
Bug 3 — populateModelDropdown() injected raw option on cold load.
Same issue: if default_model from /api/models didn't exactly match a dropdown
value, an extra option was added. Fixed by using _applyModelToDropdown()
which only selects existing options.
New helpers in ui.js:
_findModelInDropdown(modelId, sel) -- smart fuzzy match, returns matched value
_applyModelToDropdown(modelId, sel) -- sets picker, returns resolved value
Tests: 426 passed, 0 failed.
Two changes:
1. Workspace updates correctly on profile switch
switchToProfile() now applies data.default_workspace from the switch
response to the current session via /api/session/update, updates
S.session.workspace in-memory, and stores S._profileDefaultWorkspace
so the next new session also inherits the profile's workspace.
newSession() in sessions.js picks up S._profileDefaultWorkspace when
creating a new session after a profile switch.
2. Workspace chip removed from topbar
The workspace was shown in two places: the topbar chip (wsChip) AND
the sidebar bottom display (sidebarWsDisplay with name + full path).
The topbar chip was redundant, cluttered the topbar, and pushed other
chips (profile, model, clear, settings) off screen.
Removed wsChip from the topbar entirely. The sidebar display is now
the sole workspace UI, consistent and unambiguous.
Moved wsDropdown to live inside the sidebar position:relative wrapper
so it opens downward from sidebarWsDisplay. Updated the click-outside
listener to close on clicks outside sidebarWsDisplay/wsDropdown.
Removed stale wsChip update code from syncTopbar() in ui.js.
Tests: 426 passed, 0 failed.
Add full profile management to the web UI, matching the hermes-agent CLI
profile system. Profiles are isolated HERMES_HOME instances with their own
config, skills, memory, cron, and API keys.
Backend: new api/profiles.py wrapping hermes_cli.profiles, dynamic config
reloading, 5 new API endpoints, profile-aware path resolution, HERMES_HOME
env save/restore in streaming, module-level cache patching for skills_tool
and cron/jobs.
Frontend: profile chip in topbar with dropdown, Profiles sidebar panel with
CRUD UI, boot-time profile fetch, cascade refresh on switch.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- index.html: btnSend hidden by default (display:none), icon-only (upward
arrow SVG, no text label), title attribute for accessibility
- style.css: new send-btn design — 34px circle, blue fill (#7cb9ff),
subtle glow box-shadow, scale() hover/active for tactile feel,
.send-btn.visible with @keyframes send-pop-in (scale+opacity spring
using cubic-bezier(.34,1.56,.64,1) for a satisfying pop). Mobile
override updated to preserve circle dimensions.
- ui.js: updateSendBtn() — shows button with pop-in animation when
textarea has content OR files are attached and agent is not busy;
hides instantly when content is cleared. Hooked into setBusy() and
renderTray() so button state tracks all content sources correctly.
- boot.js: input event listener calls updateSendBtn() on every keystroke.
- messages.js: autoResize() calls updateSendBtn() so button disappears
immediately after send clears the textarea.
- tests/test_sprint21.py: 33 tests covering HTML structure, CSS design
(circle shape, colors, animations, keyframes), JS logic (updateSendBtn,
setBusy, renderTray, autoResize integration), and regressions
(363 total, all pass).
- File preview auto-close: clearPreview() extracted as named function
and called from loadDir(). Navigating directories (breadcrumbs, up
button, folder clicks) now automatically closes the right panel
file preview instead of leaving stale content visible.
- Thinking/reasoning display: assistant messages with structured content
arrays containing type=thinking or type=reasoning blocks now render
as collapsible gold-themed cards above the response text. Collapsed
by default, click header to expand. Works with Claude extended thinking
and o3 reasoning tokens when preserved in the message array.
- Workspace tree view (Issue #22): directories expand/collapse in-place
with toggle arrows. Single-click toggles, double-click navigates
(breadcrumb view). Subdirectory contents fetched lazily and cached.
Indentation shows nesting depth. Empty directories show "(empty)".
S._expandedDirs tracks open state, S._dirCache caches fetched entries.
Tests: 295 passed, 23 pre-existing failures, 0 regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Track A: Workspace breadcrumb navigation
- Breadcrumb path bar with clickable segments when inside subdirectories
- Up button in panel header for parent directory navigation
- S.currentDir state tracking; file ops stay in current directory
- New file/folder creation respects current subdirectory
Track B: Slash commands foundation
- New commands.js module (7th JS module) with command registry and parser
- Built-in commands: /help, /clear, /model, /workspace, /new
- Autocomplete dropdown on / input with arrow/tab/enter/escape navigation
- Unrecognized commands pass through to agent normally
Track C: Send key setting (closes#26)
- send_key added to settings defaults in api/config.py
- Settings panel dropdown: Enter (default) vs Ctrl/Cmd+Enter
- Keydown handler rewritten for autocomplete + send key preference
- Setting loaded on boot, persisted to settings.json
5 new tests, 242 total (219 passing, 22 pre-existing failures, 0 regressions).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
renderMd() now correctly renders safe inline HTML tags that AI models
emit in their responses:
Pre-pass (ui.js):
Converts <strong>, <b>, <em>, <i>, <code>, <br> to their markdown
equivalents (**text**, *text*, `text`, newline) before the pipeline
runs. Code blocks and backtick spans are stashed first so their content
is never modified.
inlineMd() helper (ui.js):
New helper for processing inline formatting inside list items,
blockquotes, and headings. Previously these used esc() directly, which
escaped <strong>/<code> tags that had already been converted from HTML
by the pre-pass — causing them to appear as literal <strong> text
instead of rendering as bold. inlineMd() applies bold/italic/code
processing and then escapes only unknown tags.
Safety net (ui.js):
After the full pipeline, any HTML tags NOT emitted by our own renderer
(i.e. <img>, <script>, <iframe>, <svg>, <object>, etc.) are escaped
via esc(). The SAFE_TAGS allowlist covers every tag the pipeline itself
produces. XSS is fully blocked.
Active session gold style (sessions.js, style.css):
Active session item now uses gold/amber (#e8a030) instead of blue,
matching the logo gradient color for better visual hierarchy.
Project color border-left is skipped when the session is active
(gold always wins). Session items get border-radius: 0 8px 8px 0
to complement the left border indicator.
Tests (tests/test_sprint16.py — 74 tests):
- Static analysis: pre-pass, SAFE_TAGS, SAFE_INLINE, inlineMd present
- Behavioural: all safe tags render in paragraphs, list items (ul+ol),
blockquotes, headings (h1/h2/h3)
- Exact screenshot regression: the 4-item list with <strong> labels
and <code> values that was showing as literal text
- XSS: 7 attack vectors blocked (<img>, <script>, <iframe>, <svg>,
<object>, XSS inside bold, XSS nested inside <strong>)
- Edge cases: code block protection, double-escaping guards, br tag,
mixed markdown+HTML, inlineMd called in list/blockquote handlers
Tests: 312 passed, 0 failed.
Session projects: named groups for organizing sessions. Project filter
bar with chips between search and session list. Create/rename/delete
projects, assign sessions via folder icon dropdown. Stored in
projects.json, project_id on Session model. 5 new API endpoints.
Code block copy button: every code block gets a Copy button in the
language header (or top-right for plain blocks). Clipboard API with
"Copied!" feedback.
Tool card expand/collapse: messages with 2+ tool cards get an
"Expand all / Collapse all" toggle above the card group.
13 new tests (237 total), all passing. No regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Restore resolve_model_provider() in _handle_chat_sync -- removed
multi-provider model routing, breaking cross-provider selection.
2. Restore new URL(path, location.origin) + credentials:include on
fetch calls -- reverted reverse-proxy auth fix from v0.16.1.
3. Revert cron import refactor (_cron_module, _real_hermes_home_env)
back to original from cron.jobs import pattern.
Tests: 201 passed, 23 pre-existing failures, 0 new regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Path traversal in _serve_static() [CRITICAL]
Sandbox resolved path to static/ directory using relative_to().
GET /static/../../../../etc/passwd now returns 404.
2. Skill category path traversal [HIGH]
Validate category param in skill save: reject values with '/' or '..'.
3. Gate /api/approval/inject_test to loopback only [HIGH]
Endpoint now returns 404 for any non-127.0.0.1 client,
preserving test functionality while blocking remote access.
4. Escape captured groups in renderMd() [HIGH]
All inline markdown regexes (bold, italic, headings, blockquote,
list items, table cells/headers, link labels) now run captured
text through esc() before inserting into innerHTML, preventing
XSS via AI-generated content.
5. SRI hashes for CDN resources + pin Mermaid version [MEDIUM]
Added integrity= + crossorigin= to all three PrismJS CDN tags.
Pinned Mermaid from floating @10 to @10.9.3 with SRI hash.
Tests: 224 passed, 0 failed.
When Hermes WebUI runs behind a reverse proxy with HTTP basic auth
(e.g. Caddy basic_auth), browsers embed credentials in the page URL.
The Fetch API and EventSource reject requests constructed from URLs
that include credentials (per Fetch spec, all modern browsers).
Fix: construct all fetch() and EventSource URLs via
new URL(path, location.origin) which strips credentials from the
base URL. Add credentials:"include" to ensure auth headers are
forwarded on each request.
