When merging PR #2105 (Hermes Run Adapter RFC) the standing concern was
that landing the RFC unconfirmed would invite the speculative-fragment
implementation pattern we just had to put on hold with PR #2071 — well-
written 651-LOC standalone scripts with no callers.
Add a single bullet to the conventions block so the contract is explicit:
an RFC is a design direction, not an invitation to PR fragments against
it. Implementation slices need maintainer confirmation first.
Applied during stage-341 build, not requested from @Michaelyklam — the
guardrail belongs in the conventions doc itself rather than as a one-off
ask on this PR.
Two follow-ups from Opus pre-release review of stage-336:
1. tests/conftest.py — autouse session fixture that removes
HERMES_WEBUI_SKIP_ONBOARDING from os.environ for the whole pytest run, and
restores it after. Hosting providers and isolated harnesses set this var
to short-circuit the onboarding wizard, but it leaked into pytest and
caused tests that exercise apply_onboarding_setup() to fail with cryptic
FileNotFoundError. Tests that specifically validate the short-circuit
behavior can opt back in with monkeypatch.setenv. Surgical per-test
delenv calls remain as defense-in-depth but are now redundant.
2. docs/rfcs/README.md — one-line note that first-time contributor RFCs
should be discussed in an issue before opening a PR. Gates drive-by
design-doc PRs without us having to decline them on contribution.
Verified: 96 onboarding-related tests pass with HERMES_WEBUI_SKIP_ONBOARDING=1
exported in the test runner env (would have failed before this fixture).
Moves docs/turn-journal-rfc.md → docs/rfcs/turn-journal.md, establishing
the convention for future design documents on hermes-webui's data-at-rest
and recovery surfaces. Adds docs/rfcs/README.md describing when an RFC
applies (large changes, durability/recovery semantics, new infrastructure
primitives) and the simple status header convention.
Polish on turn-journal.md:
- Added 3-line status header (Status / Author / Created) at top.
- Light tone edits on two flourishes that read fine in a PR description
but felt off in permanent repo documentation. Author's voice preserved
throughout the rest of the document.
Co-authored-by: ai-ag2026 <261867348+ai-ag2026@users.noreply.github.com>
Note: PR #1827 was branched before v0.51.19 shipped #1812, which
introduced an initial (pure live-fetch) Codex provider card hook in
api/providers.py at the same line range. The contributor's PR was
filed AFTER #1812 shipped but their diff didn't yet account for it.
Stage 314 absorbs the contributor's intent (visible Codex cache
merge for gpt-5.3-codex-spark visibility) by replacing the v0.51.19
hook with the richer merged version directly in stage. Production
code change ≡ what the contributor's PR would have produced if
rebased onto current master. Test file + pr-media adopted verbatim.
Marker commit so the stage log makes the absorption visible.
Closes#1695.
@Patrick-81 reported the bare "AIAgent not available -- check that
hermes-agent is on sys.path" error on a symlinked install (~/Programmes/hermes-agent
linked to ~/hermes-agent). The maintainer's response — three diagnostic
commands plus `pip install -e .` in the agent dir — fixed it for them.
This PR captures both halves of that learning so the next user with the
same shape doesn't have to file a new issue:
1. **Error message diagnostic block.** New helper
`_aiagent_import_error_detail()` in api/streaming.py builds a multi-line
diagnostic when the import fails, including:
- the running Python interpreter
- HERMES_WEBUI_AGENT_DIR (set value, or "(not set)")
- sys.path entries that mention hermes/agent (or "no entries mention..."
— itself a strong diagnostic signal)
- the most-common fix (`pip install -e .` in the agent dir)
- a pointer to docs/troubleshooting.md
The original error message string is preserved as the FIRST line so
existing log scrapers and docs-search keep matching.
Helper is kept as a separate function so it stays out of the hot path
until we actually need to raise — building it on every successful import
would be wasted work.
2. **New docs/troubleshooting.md.** Symptom → Why → Diagnostic commands →
Fix → When-to-file-a-bug template, with one entry to start: the
"AIAgent not available" flow Patrick-81 walked through. Future
recurring failure modes follow the same template. Required a one-line
addition to .gitignore — docs/* is gitignored with an allowlist, and
the new file needed `!docs/troubleshooting.md` to be tracked.
3. **README link.** docs/troubleshooting.md added to the `## Docs` section
so users know where to look first.
13 regression tests in tests/test_1695_aiagent_import_error_detail.py:
9 for the helper output shape (preserves original message line, includes
running python, shows HERMES_WEBUI_AGENT_DIR set/unset both ways, includes
pip-install-e hint, points at troubleshooting doc, lists relevant sys.path
entries when present, says "no entries..." when absent, output is multi-line)
plus 4 for the docs-presence regression (file exists, has the AIAgent
section, includes pip install -e ., describes the diagnostic chain with
readlink + agent/__init__.py verification).
190 streaming/aiagent tests pass after the change. ast.parse on
api/streaming.py clean.
CI failure on prior push was due to the docs/* gitignore swallowing the
new troubleshooting.md file silently — this commit adds the allowlist
entry so the file is tracked.
Adopt the UI media from @Michaelyklam's parallel-discovery PR #1641 which
shipped the same one-character regex relax fix for #1618. PR #1641 is
being closed as superseded by #1642 (which carries nesquena APPROVED +
322 LOC test suite); preserving Michael's UI evidence here so the visual
proof of the fix lives in-tree alongside the canonical PR.
Co-authored-by: Michael Lam <Michaelyklam1@gmail.com>
Three changes from the pre-merge Opus review:
**MUST-FIX** — XPC_SERVICE_NAME false-positive on macOS Terminal
macOS launchd sets `XPC_SERVICE_NAME` in EVERY Terminal-spawned shell, not
just real services. Typical noise values: `"0"` (truthy in Python!) and
`"application.com.apple.Terminal.<UUID>"`. A bare `os.environ.get(name)`
existence check would auto-promote interactive `./start.sh` runs to
foreground mode on every Mac dev machine — silently breaking the most
common installation path (no /health probe, no browser open, no log file,
hanging shell).
Fix: new `_is_real_supervisor_value()` helper that filters noise. For
`XPC_SERVICE_NAME` specifically, reject `"0"` and any `"application.*"`
prefix. Real launchd plists use reverse-DNS Label form (`com.<rdns>.<svc>`)
which still triggers correctly.
7 new tests in `TestXPCServiceNameNoiseFilter`:
- 4 noise values (`0`, Terminal.app, iTerm2, VSCode) → no detection
- 3 real Label forms → correct detection
- Mixed env with XPC noise + real INVOCATION_ID → falls through to systemd
**SHOULD-FIX 1** — Test env leakage
The original `clean_env` fixture stripped supervisor-detection env vars
but not the resolved bootstrap vars (HERMES_WEBUI_HOST/PORT/AGENT_DIR)
that `main()` mutates onto `os.environ`. After
`test_foreground_exports_resolved_env_vars` ran, later tests would import
bootstrap with polluted defaults (DEFAULT_HOST="0.0.0.0" instead of
"127.0.0.1"). Existing assertions still passed (tautological vs DEFAULT_*),
but it was a footgun for future tests.
Fix: extend `clean_env` to also `delenv` the three resolved vars before
each test.
**SHOULD-FIX 2** — Pre-execv executability guard
If `discover_launcher_python` returns a path that doesn't exist or isn't
executable, `os.execv` raises OSError → wrapper catches → SystemExit(1)
→ supervisor restarts → loop forever. That's exactly the failure mode
this PR is supposed to eliminate.
Fix: `os.access(python_exe, os.X_OK)` check before execv. Converts
infinite supervisor loop into a single visible RuntimeError.
1 new test in `TestForegroundExecutabilityGuard` pinning that the guard
fires before execv when the python path is non-executable.
**Docs** — supervisor.md updates
- New section explaining the XPC_SERVICE_NAME noise filter and what
values trigger / don't trigger detection
- New section listing supervisors that are NOT auto-detected (runit,
daemontools, PM2, Foreman/Honcho, custom shell-script supervisors)
with explicit recommendation to set HERMES_WEBUI_FOREGROUND=1
Verification
- 3820 tests pass (+9 from this commit's new tests vs the original PR
push of 3811)
- Filter manually verified end-to-end with the live os.environ:
XPC=0 → None, XPC=application.* → None, XPC=com.example.foo → triggers
- run-browser-tests.sh ALL CHECKS PASSED on the worktree
Items deferred from the Opus review
- #4 chdir target may not exist: REPO_ROOT comes from __file__.resolve()
so it's stable; not a real concern in practice
- #6 two startup messages in foreground mode: cosmetic, useful for
diagnostics
- #7 stricter explicit-only mode: leaves user the override of just not
passing --foreground (current behavior)
- #8 test stub return value: trivial, can fix later if regression surface
- #9 argparse positional-after-option ordering: test reads fine
These can be follow-up issues if anyone hits them.
Issue #1458 reports persistent-host crashes (≥1/day) when running the WebUI
under launchd KeepAlive on macOS. Root cause: `bootstrap.py` calls
`subprocess.Popen([python, "server.py"], start_new_session=True)`, probes
/health, then exits 0. Under any process supervisor (launchd, systemd,
supervisord, runit, s6), the supervisor sees its tracked PID exit, marks
the program as "completed," and respawns it. The new bootstrap fails to
bind port 8787 (orphaned server still has it), exits non-zero, supervisor
respawns again — loop until the orphan crashes for some other reason and
the next respawn finds the port free.
This PR addresses Bug #1 of the three failure modes tracked in #1458:
the `bootstrap.py` double-fork breaking process supervisors. Bug #2
(state.db FD leak) and Bug #3 (HTTP-unhealthy wedge) remain open under
the same issue — they need diagnosis data before a fix can land.
Changes
-------
1. `bootstrap.py`:
- New `--foreground` argparse flag with help text mentioning launchd /
systemd / supervisord.
- New `_detect_supervisor()` that returns the env var name for any
supervisor it detects: `INVOCATION_ID` / `JOURNAL_STREAM` /
`NOTIFY_SOCKET` (systemd, s6), `XPC_SERVICE_NAME` (launchd),
`SUPERVISOR_ENABLED` (supervisord), or `HERMES_WEBUI_FOREGROUND` for
the explicit user opt-in. Truthy values for the explicit opt-in:
`1` / `true` / `yes` / `on` (case-insensitive).
- `main()` branches on `args.foreground or _detect_supervisor()`:
- **Foreground path:** chdir to `agent_dir or REPO_ROOT`, then
`os.execv(python, [python, server_path])` to replace the bootstrap
process image with the server. The supervisor sees the long-lived
server as the original child. No `wait_for_health` probe — the
supervisor's KeepAlive / Restart=on-failure handles liveness.
- **Default path:** unchanged. Spawn server as detached child via
`Popen + start_new_session=True`, probe /health, return 0. This
still works for interactive `bash start.sh` invocations.
- Resolved env vars (HOST/PORT/STATE_DIR/AGENT_DIR) are now mutated on
`os.environ` directly instead of into a local `env` copy so they
are inherited across `os.execv`.
2. `docs/supervisor.md` (new): runnable launchd plist, systemd .service,
and supervisord conf examples + a diagnostic recipe (`lsof` + ppid
chain) for catching the orphan-loop in production.
3. `.gitignore`: allowlist `docs/supervisor.md` (the directory uses an
opt-in pattern; matches the existing `!docs/docker.md` precedent).
4. `tests/test_bootstrap_foreground.py` (new): 35 regression tests
covering the argparse flag, `_detect_supervisor()` behavior across all
five supervisor env vars, the explicit opt-in's truthy/falsy values,
and `main()`'s execv-vs-Popen routing decision under each input
combination. `os.execv` is monkeypatched in the routing tests — we
pin the structural choice (which call is made, with which args, in
which cwd, with which env) not the post-exec behavior.
Why this scope and no more
--------------------------
Bug #2 (state.db FD leak) lists 5 candidate paths and asks the reporter
for `lsof -p <pid> | sort | uniq -c | sort -rn | head -20` output to
disambiguate. Until that data lands, any "fix" would be speculative —
explicitly out of scope per the contributor-pickup comment on the issue.
Bug #3 (launchd-running, port-listening, HTTP-unhealthy) was added in
@stefanpieter's reply comment. Diagnosis is in flight; no concrete fix
shape yet. Also out of scope.
Running locally end-to-end verifies the behavior:
```
[bootstrap] Starting Hermes Web UI on http://127.0.0.1:8789 (foreground mode: --foreground)
$ pgrep -af 'server.py'
2997632 /home/.../python /tmp/wt-fix-1458/server.py
$ ps -o ppid -p 2997632
2997581 ← bash that ran bootstrap.py — same PID as the original bootstrap
$ ps -p 2997581 -o cmd
... bootstrap.py ... ← but exec'd into server.py
```
The same PID that bash forked for `bootstrap.py` is now `server.py`.
A supervisor watching that PID would correctly observe the long-lived
server. No double-fork.
Verification
------------
- 3811 tests pass (`pytest tests/` — full suite, +51 from this PR plus
master-merge-in)
- All 35 new bootstrap-foreground tests pass
- `bash scripts/run-browser-tests.sh` PASS (HTTP API checks against worktree)
- `bash scripts/webui_qa_agent.sh 8789` PASS (23/23 visual QA)
- Live verified: server starts cleanly under both `--foreground` and
`HERMES_WEBUI_FOREGROUND=1`; PID lineage confirms no double-fork
Closes#1458 (Bug #1 only). Bugs #2 and #3 remain tracked under the
issue.
REQUIRED:
- _fully_unquote_path range(3) -> range(10) — defense-in-depth so quadruple-
encoded .. is rejected by validator instead of slipping through (not
exploitable but contract violation)
- docs/EXTENSIONS.md trust-model callout moved to top of file with explicit
'don't enable in untrusted env / don't point at user-writable dir' guidance
NICE-TO-HAVE (taken since Nathan asked for all fixes big and small):
- URL list cap at _MAX_URL_LIST=32 to avoid pathological rendering
- One-shot WARNING log for rejected URLs (silent drop now visible to admin)
- One-shot WARNING log for URL list truncation
- MIME map: ttf (font/ttf), otf (font/otf), wasm (application/wasm)
5 regression tests in tests/test_pr1445_opus_followups.py pin all invariants.
Combines PR #1428 (UID/GID alignment) with a broader Docker reliability pass
that addresses recurring user reports about compose files not working.
Constituent PR:
- #1428 sunnysktsang - Align agent UID/GID with webui (fixes#1399).
Two- and three-container compose files had agent at UID 10000 (image
default) and webui at UID 1000 (WANTED_UID default), causing permission
denied on shared hermes-home volume. All services now use ${UID:-1000}.
Plus broader Docker UX overhaul:
- All 3 compose files document HERMES_SKIP_CHMOD/HERMES_HOME_MODE escape
hatches inline (the v0.50.254 fix wasn't surfaced for Docker users).
- New .env.docker.example template covering UID/GID, paths, password,
permission handling. UID/GID are uncommented with placeholder values
per Opus advisor (so macOS users don't skim past).
- New docs/docker.md - comprehensive guide: 5-min quickstart, failure
mode table with one-line fixes, bind-mount migration, multi-container
architecture diagram, macOS Docker Desktop VirtioFS note, link to
community sunnysktsang/hermes-suite all-in-one image.
- README Docker section rewritten - clearer quickstart, failure-mode
table, link to docs/docker.md. Stale /root/.hermes references removed.
Plus Opus pre-release advisor MUST-FIX:
- HERMES_HOME_MODE has DIFFERENT semantics in the WebUI vs the agent
image. WebUI: credential-file mode threshold (0640 allows group bits).
Agent: HERMES_HOME directory mode (default 0700). 0640 on a directory
has no owner-execute bit, so the agent can't traverse its own home and
bricks. My initial draft recommended HERMES_HOME_MODE=0640 in agent
service blocks - corrected to 0750 across all 4 surfaces (compose
files, .env.docker.example, docs/docker.md). 3 regression tests pin
the asymmetry.
12 regression tests total in test_v050260_docker_invariants.py.
Full suite: 3627 passed, 0 failed.
Nathan explicitly authorized merge with my own review + Opus only, no
independent review needed.
When a user switched profiles and created a new session, the session
was saved to the default profile directory instead of the active
profile directory — because get_hermes_home_for_profile() silently
fell back to _DEFAULT_HERMES_HOME when the profile directory didn't
exist yet on disk.
Root cause: api/profiles.py:156 had `if profile_dir.is_dir(): return
profile_dir; return _DEFAULT_HERMES_HOME`. New profiles (no session
yet, so no dir) routed every session back to default.
Fix: remove the is_dir() guard, return the profile path
unconditionally. The profile directory is created on first use by
the agent/session layer.
5 regression tests in tests/test_issue1195_session_profile_routing.py:
existing-profile, non-existent-profile (the core fix), None, empty-
string, 'default' all return the expected path.
Co-authored-by: bergeouss <bergeouss@users.noreply.github.com>
Merged as v0.50.226.
Integration branch absorbed @aronprins's original PR #1141 with one reviewer fix from @nesquena (`1d11646`: queue hide tooltip updated to reference the queue pill, not the removed titlebar badge).
**Full gate results:**
- 2595 tests passing ✅
- Browser QA 21/21 (desktop 1440×900 + mobile iPhone 14) ✅
- Independent review: APPROVED by @nesquena ✅
Thank you @aronprins for the clean PR — the titlebar is properly restored.
Removes the bubble_layout toggle from Settings, all persistence, CSS, i18n strings, and the UI docs demo. The CSS was already effectively dead. Users with a saved bubble_layout value in settings.json get a clean migration via _SETTINGS_LEGACY_DROP_KEYS.
Credit: @aronprins (PR #760 / #777)
Co-authored-by: aronprins <aronprins@users.noreply.github.com>
* docs: add HERMES.md deep-dive, Why Hermes section in README, and screenshot layout
- HERMES.md: full why-Hermes document -- assistant vs. agent mental
model, three pillars (memory/scheduling/reach), four-category
taxonomy of AI tools, per-tool comparison sections with tables
(Claude Code, Codex CLI, OpenCode, Cursor/Copilot, Claude.ai),
compounding advantage, who it's for, what it's not, quick reference
- README: hero screenshot stays full-width; two new UI screenshots in
side-by-side HTML table with captions below
- README: new Why Hermes section with 6-bullet summary, comparison
table, and link to HERMES.md
- README: HERMES.md added to Docs section
- docs/images/: two UI screenshots (workspace browser, sessions view)
* docs: fact-check and update all comparisons; add Open Interpreter section
Researched current state of each tool before updating:
Claude Code:
- Scheduled jobs: now Partial (has /loop session-scoped, cloud-managed
/schedule via claude.ai/code, and desktop app automations); updated
table to reflect this with footnotes distinguishing self-hosted cron
- Persistent memory: Partial (CLAUDE.md, MEMORY.md, rolling auto-memory
but not full automatic cross-session recall)
- Provider-agnostic: No -- supports Bedrock/Vertex but Claude models only
- Web UI: Yes but Anthropic-hosted (not self-hosted)
Codex CLI:
- Persistent memory: Partial (session history + AGENTS.md since v0.100.0)
- Scheduled jobs: Partial (desktop app Automations only; CLI has no native
scheduling as of early 2026, open feature request)
- Provider-agnostic: Yes (10+ providers)
OpenCode:
- Web UI: now Yes (embedded in binary + official desktop app)
- Persistent memory: Partial (SQLite sessions + AGENTS.md, not semantic)
- Messaging: community Telegram bot only, not first-party
Open Interpreter: added as new comparison section
- Most common 'why not just use this' question; addressed head-on
- Session-scoped, no persistent memory by their own docs, no scheduler,
no messaging integration; powerful for one-shot tasks, not always-on
README Why Hermes table: updated to include Open Interpreter column,
fixed Claude Code self-hosted row (No -- scheduling runs on Anthropic
cloud), added footnotes for partial entries
* docs: add OpenClaw comparison; update category framework and quick reference table
OpenClaw (openclaw.ai, MIT, 347k stars) is the most direct Hermes
competitor -- both are open-source, self-hosted, always-on agents with
persistent memory, cron, and messaging integration. Added:
- Full OpenClaw section in HERMES.md with honest comparison: where it
wins (15+ messaging platforms incl. iMessage/WeChat, native Chrome CDP
browser control, voice wake words, ClawHub marketplace) and where
Hermes differs (self-improving skills system, Python/ML ecosystem,
web UI, multi-profile, sub-agent orchestration)
- Category 4 framework updated: now lists both Hermes and OpenClaw,
with the key architectural distinction called out
- Quick reference table expanded to include OpenClaw column (now 8 tools)
- New rows added: self-improving skills, browser/computer control,
Python/ML ecosystem
- README Why Hermes table updated: OpenClaw replaces OpenCode column,
self-improving skills row replaces generic skills row, callout line
at bottom addresses OpenClaw head-on
* docs: major accuracy pass -- OpenClaw deep-dive, Claude Code corrections, drop Open Interpreter
OpenClaw:
- Expanded comparison from a table to a full prose section with
'Where OpenClaw wins' / 'Where Hermes wins' structure
- Honest about OpenClaw strengths: 15+ messaging platforms, native
Chrome CDP browser control, voice wake words, 13k+ ClawHub skills
- Hermes advantages called out clearly: self-improving skills as a
first-class automatic loop (vs marketplace-install model), stability
(documented OpenClaw update regressions, Telegram breakage in early
2026, WhatsApp protocol instability), security (156 CVEs and 1,184
malicious skills found in ClawHub audit vs Hermes's no marketplace
attack surface), Python/ML ecosystem, full web UI vs dashboard-only,
and first-class multi-profile support
- Category 4 framework updated to name both Hermes and OpenClaw
- Table updated: added stability/security rows, corrected web UI row
(OpenClaw has a gateway dashboard but not a full chat UI)
Claude Code corrections (researched against official docs at code.claude.com):
- Skills/Hooks: changed from No to Yes -- has a full Hooks system (13
event types, 4 handler types) and a Plugin/Skills marketplace since
v2.0.12; unified with slash commands in v2.1.0
- Messaging: changed from No to Partial -- Channels feature (Telegram,
Discord, iMessage, Webhooks) in research preview since v2.1.80; deep
Slack integration that triggers cloud sessions and creates PRs
- Added Claude Cowork row: separate product with 38+ connectors
(Slack, Gmail, Teams, Notion, Jira, Salesforce, etc.)
- Scheduling footnote updated: cloud-managed has 1-hour minimum interval
- Provider-agnostic clarified: routes through Bedrock/Vertex but always
Claude models; cannot swap to GPT or Gemini
Open Interpreter removed:
- Less relevant comparison than OpenClaw for the 'always-on agent' frame
- Kept coverage focused on the tools people actually compare Hermes to
Quick reference table:
- Now 7 tools wide (added OpenClaw, kept Claude Code, Codex, OpenCode,
Cursor, Claude.ai, Hermes)
- New rows: self-improving skills, browser/computer control, stability
- Updated: Claude Code messaging to Partial, OpenClaw web UI to
'Dashboard only', skills rows differentiated by type
* docs: apply full editorial pass from hermes-edit-list.md
Writing patterns fixed:
- Em dashes reduced by ~80%; replaced with commas, periods, parens
- All 'Not X, it's Y' negative parallelism rewritten as positive
statements; 'What Hermes Is Not' section renamed 'Scope and Limits'
and reframed positively throughout
- 'It compounds.' standalone flourish removed
- 'meaningfully' removed everywhere (was appearing 3+ times)
- 'leverages' -> 'uses' in README
- 'remembers everything' softened to 'retains context across sessions'
- Bolded Hermes column in Quick Reference table un-bolded (only genuine
differentiator cells kept bold: self-improving skills, always-on,
orchestrates other agents)
- 'The honest summary' framing removed from OpenClaw section
- 'Hermes is different.' cliche transition cut from README
- Rule-of-three slogans trimmed (e.g. 'Same agent, same memory...')
- 'tired of re-explaining' -> 'don't want to re-explaining'
Duplicate content removed:
- 'day one / day one hundred' comparison kept only in Compounding
Advantage section; removed from Pillar 1
Factual accuracy fixes:
- Claude.ai comparison updated: memory now auto-generated from history
(not just user-curated); code execution and file read/write noted
as sandboxed (Artifacts), not flat No
- Category 2: Windsurf framed as 'earliest' on memory, Copilot
'catching up'; removed overconfident 'most mature' claim
- Category 4 qualifier: 'as of early 2026' added
- '1-hour minimum' for Claude Code cloud scheduling softened to
'minimum interval applies' (specific claim unverified)
- Claude Code scheduling table note: 'cloud or desktop-app only'
(was just 'cloud-managed or session-scoped')
- README claim 'No other open-source tool combines...' removed;
was false because OpenClaw does combine all three
- OpenClaw self-improving skills: 'No' -> 'Partial' with clarification
- README OpenClaw callout: 'relies on a marketplace' softened to
'skill system centers on a community marketplace'
- 'meaningfully more stable' -> 'more stable'; 'supply chain issues'
-> 'security incidents involving malicious skills'
- OpenClaw star count: '347k+' -> '~347k' (moving fast)
- Stability row added to OpenClaw table; bold removed from table
---------
Co-authored-by: Hermes <hermes@localhost>
