test-convert-outputs.sh's Python fence model read fences only at column 0. With an indented opening fence and a column-0 closer - valid CommonMark that GitHub renders, lint accepts and convert_openclaw splits correctly - the model missed the opener and treated the closer as its own opener, so it reported a torn fenced block across SOUL.md/AGENTS.md and failed the Check Tools workflow on an agent file that was fine.
Mirror lib.sh's fence_open_p / fence_closes_p (up to three spaces, same character, a closer at least as long as the opener with nothing but whitespace after the run) so the eval sees the same blocks the converter does.
test-convert-fence-model.sh builds a throwaway repo with one such fixture and runs the real eval; on the previous model it fails with the false torn-block report, and with this change all tools validate. The full roster still passes with zero manifest drift.
Exactly the 20 agents the 20 merged fixes touch; no other agent or tool line moves.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(hermes): match config items with inline comments
The plugins.enabled / plugins.disabled item scan compared raw line text, so
a YAML inline comment became part of the value it matched against. Three
concrete failures followed on configs users can edit by hand:
- an existing enabled entry carrying a comment ("agency-agents-router # x")
was not recognized, so the plugin was appended a second time;
- a stale disabled entry carrying a comment was not removed, so the
installer reported "enabled plugin ... in config.yaml" while the plugin
stayed disabled;
- the corrupted-glue repair counted "- " in the whole line, so a comment
containing " - " was split into fake list entries: the installer silently
rewrote a comment such as "# keep - rotate this one first" into two
enabled entries.
Strip a trailing YAML comment ("#" only starts one after whitespace) before
matching, counting, and splitting, and use the same value helper in the
post-repair re-check so a quoted entry is recognized too. No config format
or install behavior change beyond those shapes.
check-hermes-config-rewrite.py gains four cases for them plus two general
invariants: the plugin appears exactly once, and the writer may not gain or
lose unrelated enabled entries outside the documented glue repair. The
suite fails on the previous code with exactly those four cases and passes
with this patch.
* test(hermes): pin a quoted entry that also carries a comment
Independent verification on the PR (phant0um) covered an existing quoted
entry with an inline comment (`- "agency-agents-router" # quoted`). The fix
recognizes it through the comment-stripped item value; pin that shape so it
cannot regress. On the previous code the extended suite fails with this case
plus the original four; with the fix all 21 cases pass.
fence_closes_p() required a closing fence's indent to be <= the opener's.
That is not a CommonMark rule and not how GitHub renders: a closing fence
may be indented up to three spaces regardless of the opener. A GitHub-valid
document was therefore read as still open, so lint reported a false
"does not nest" (and a false "never closed") and the OpenClaw split kept the
next "##" heading inside the previous section, leaving AGENTS.md empty.
The closing rule is now exactly the four conditions already enforced: same
character, run at least as long as the opener, indent 0-3, and nothing but
whitespace after the run. Callers still pass the opener's indent as the
optional fourth argument; it is documented as deliberately not consulted.
Regression fixtures in test-lint-fences.sh pin the helper boundaries (2- and
3-space closers close; a shorter or 4-space-indented run does not; tilde
closers follow the same rule), a lint run on the GitHub-valid document, and
the OpenClaw split. They fail on the previous code at the first new
assertion: "fence_closes_p rejected a bare closer indented 2 spaces under
an unindented opener".
Exactly the 43 agents the 49 merged fixes touch; no other agent or tool line moves.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
From @Hotragn. 17 agents list Claude Code tool names (Read, Write, Bash...). Qwen Code resolves subagent tools by exact name or display name, so those three matched nothing and were never granted. Traced in Qwen Code 0.24.7: resolveToolNames preserves unknown names, and getTool() is a plain lookup with no alias translation. Now mapped to read_file, write_file and run_shell_command; other names pass through unchanged. The manifest moves exactly those 17 agents.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
From @nolanchic. ServiceNow platform developer and step-by-step troubleshooter: Business Rules, Script Includes, GlideRecord/GlideAggregate, GlideAjax, ACLs, Flow Designer. Maintainer follow-up: the GlideAjax example now reads its parameter with this.getParameter() (the function argument never receives it), and an unverified system-property check is replaced with Transform Map guidance.
Closes#250
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
From @Subhodip-Chatterjee's Developer Relations submission. DX Engineer joins product/ (onboarding friction, SDK ergonomics, error messages, first-run experience). Community Builder joins marketing/ as Developer Community Builder (community architecture, contributor programs, health metrics). Developer Advocate and Docs Engineer were not added, because specialized/Developer Advocate and engineering/Technical Writer already cover those roles.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With a spaced --path, the parallel workers get the path split into words
and reject the stray words as unknown options (the bug #755 fixes). Unknown
options used to exit 0, so every worker "succeeded" having installed
nothing, and this case's exit-code assert passed while the count assert
next to it was already an xfail for #755.
Now that unknown options exit 1, the parent exits non-zero there, which is
the truth. The exit-code assert becomes an xfail for #755 like its
neighbour; both will report "appears to have landed" once #755 merges.
Both scripts handled an unknown option with `error ...; usage`, and usage()
always ended in `exit 0`:
$ ./scripts/convert.sh --tol codex; echo $?
[ERR] Unknown option: --tol
...usage...
0
so a mistyped flag in CI or a wrapper script read as success. usage() now
takes a status. --help still prints to stdout and exits 0; an unknown
option prints the usage text to stderr and exits 1, the same as a bad
--tool or --division value already does.
convert.sh --help also printed a hard-coded `sed -n '3,28p'`. The option
notes live on lines 33-34, so --parallel and --jobs never appeared, and
--out was only in the synopsis. It now uses USAGE-START/END sentinels, as
install.sh has for a while, and the option list covers --tool and --out.
test-cli-usage.sh checks both scripts: an unknown option exits non-zero
with its name and the usage on stderr and nothing on stdout, --help exits 0
without printing the sentinels, and convert.sh --help describes --tool,
--out, --parallel and --jobs. It fails on main at the first check and runs
in check-tools.yml.
strategy/runbooks.json names agents by file stem, the corpus id the desktop
app maps to catalog entries ("engineering-frontend-developer"). The installer
only accepted the slug built from `name:` ("frontend-developer"), and for
206 of the 279 agents those differ. Of the 36 agents the four runbooks list,
35 could not be selected by the ids the runbooks give them:
$ install.sh --tool claude-code --agent engineering-frontend-developer --dry-run
[ERR] Unknown agent 'engineering-frontend-developer'. Use --list agents ...
So there was no way to install a runbook's team from the CLI without
translating every id by hand.
resolve_agent now takes a slug, a display name (as before, via slugify), or a
file stem, and returns the install slug. Slugs are tried first. No stem
equals a different agent's slug today, and stems are unique, so nothing that
resolved before resolves differently now.
Resolution also stops rescanning the roster per request. agent_slug_exists
ran get_field over all 279 files for each requested agent; roster_index
builds the slug/stem table once in the parent shell. A 36-line agents file
went from ~10,000 get_field calls to 279.
test-agent-selection.sh gains a stem case and feeds each runbook's roster to
--agents-file, checking that it resolves to exactly its own agent count. It
fails on main at the first runbook. The usage text and the README selection
section say which ids are accepted, with an example of installing a
runbook's team.
A sequential install called install_tool bare under set -e, so the first
tool that returned 1 exited the whole script. With copilot, cursor, qwen,
codex and hermes detected and no integrations/cursor:
[1/5] copilot [OK]
[2/5] cursor [ERR] integrations/cursor missing. Run convert.sh first.
(exit 1)
qwen, codex and hermes were never attempted, and nothing said so. The same
happens for any ensure_converted failure or an OpenClaw registration error.
Each tool now runs in a subshell with errexit on. It is not
`install_tool "$t" || failed+=(...)`: bash ignores errexit for everything on
the left of ||, subshells included, so a failing cp inside a tool would carry
on as if it had worked. Instead the parent turns errexit off for that one
command, and the subshell turns it back on for itself.
A failed tool is recorded and the loop moves on. The done box reads
"Installed 2 of 3 tool(s)" instead of "Done!", and the run ends with
"Failed: cursor — see the [ERR] line under each above" and exit 1. A clean
install is unchanged. Nothing in install_* sets globals the parent reads
afterwards, and the SKIPPED_LOG cleanup trap does not fire in a ( )
subshell, so running tools in a subshell loses nothing.
Parallel mode is untouched; its log replay on failure is item 1 of #917.
test-install-continue-after-failure.sh builds a fixture repo with
gemini-cli and qwen output and no cursor output, installs all three, and
checks that qwen still installs, cursor's error is shown, the summary names
cursor, and the exit is non-zero. It also checks that a clean run still
exits 0. It fails on main at "qwen ... was never installed" and runs in
check-tools.yml.
path_collision_group put claude-code and copilot in their own group on the
assumption that they write <division>-<slug>.md. They copy the source file
under its own name, and 73 of the 279 agents are named <slug>.md already:
all 20 in game-development/, 38 in specialized/, 6 in spatial-computing/ and
9 elsewhere. For those, the file is the same one gemini-cli, opencode, qwen
and zcode write.
$ install.sh --tool claude-code,qwen --agent game-designer --path out --no-convert
[OK] Claude Code: 1 agents -> out
[OK] Qwen Code: installed 1 agents to out
$ head -2 out/game-designer.md
---
name: 'game-designer' <- qwen's frontmatter; the Claude Code file is gone
Both tools report success and the install exits 0. A full install of any of
those pairs into one --path loses 73 files the same way.
The six tools are now one group. That also refuses a pair like claude-code
and gemini-cli for a selection with no slug-named agent in it, which could
technically share a directory; the guard works per tool pair, and a full
install collides on 73 files, so treating the pair as colliding is the
honest answer. claude-code with codex, cursor, kimi and the other
distinct-output tools still share a --path as before.
test-install.sh picks a slug-named agent from game-development/ and checks
that claude-code,qwen into one --path is refused and writes nothing. The
refusal comes before conversion, so the case stays cheap.
Markdown has no nesting at equal fence length. Inside a ```markdown
template, a ```bash line is text, and the example's bare ``` closes the
template. Seven agents nest example blocks inside a template that way, so on
GitHub the rest of each template renders as headings and prose, and in three
of them the last block then runs open to the end of the file:
design-ux-architect code from line 414 to EOF
ats-validator-architect code from line 307 to EOF
master-plan-architect code from line 102 to EOF
technical-writer, incident-response-commander, developer-advocate,
workflow-architect templates end early, rest renders as markdown
The fix for each is the one CommonMark intends: fence the outer template
with four backticks so the three-backtick examples stay inside it. Nine outer
blocks change; no content does.
Nothing caught this because lib.sh's fence_closes_p accepted "```python" as
a closing fence. CommonMark allows only whitespace after a closing run, and
GitHub renders by that rule. With the old rule the linter and the OpenClaw
split read those seven files one way and GitHub another, so each agreed with
itself and all three disagreed with what users see. fence_closes_p now
requires a bare run, and the OpenClaw split invariant in
test-convert-outputs.sh uses the same rule.
lint-agents.sh gets two errors, each with the file line number:
- a fence line inside an open block that is long enough to close it but
carries an info string -- the only way that happens is an attempted
nest, and the message says how to fix it
- a block that is still open at the end of the file
The body loop now reads the file from the line after the frontmatter
instead of from $body, which drops every "---" line and so cannot give a
line number. It also keeps a last line with no trailing newline; two agents
end on a closing fence with none, and without that they read as unclosed.
The whole roster lints with 0 errors after the fix; on main the same check
reports 18 errors across the seven files.
scripts/test-lint-fences.sh covers both errors, the suggested fix, tilde
fences, and fence_closes_p directly, and runs in check-tools.yml.
#915 made registration failures visible (main swallowed them with `|| true` and
reported success). But it returned at the first failure, so every agent after it
went unregistered. In real OpenClaw 2026.9.6 with one registration forced to fail,
batch 2 registered only 1 of the 2 agents that could have succeeded.
Keep registering the rest, name every failure in one message at the end, and
still exit non-zero. Verified against real OpenClaw (node:24, official npm
package): the failure case now registers both good agents and exits 1; the
normal install and the re-run are unchanged. #915's own test still passes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
#916 stops cp from writing through a symlink at an agent's destination, which on
main overwrites whatever the link points at (verified: a user's file was replaced
by an agent). But it made one stray link fatal: under `set -e` the refusal aborted
the install at the first link, leaving a partial roster (57 of 279 in testing).
It also refused our own links, so re-installing without --link after a --link
install failed for every file (main fails there too, with "are the same file").
- A link into this clone ($REPO_ROOT) is ours: remove it and copy. That's the
intended --link -> copy switch, and the clone's sources stay untouched.
- Any other link is skipped with a warning, recorded, and the install continues.
The final summary lists what was not installed. A temp file carries the list so
it survives the parallel installer's subshells.
- test-install.sh: #916's case now asserts skip-and-continue (sentinel intact,
link left in place, the rest of the selection installs, summary reports it);
new case for --link then copy (link replaced by a real file, source unchanged).
65 passed, 0 failed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
`rm -rf "$dest/"` follows a symlink and empties its target. basename ignores a
trailing slash, so HERMES_PLUGIN_DIR=".../agency-agents-router/" passed every
check, and a symlinked destination (our own --link install, or a user's folder)
had its target's contents deleted. #903's ownership check didn't cover it: the
check reads through the link and the rm still follows it.
- Strip all trailing slashes from the destination before any check.
- A symlinked destination is removed with `rm -f` (the link itself), never
`rm -rf`.
- test-install-hermes-destination.sh: two new cases (trailing slash on a
symlink to a folder with a matching plugin.yaml; re-install over a --link
install). Both fail before this commit and pass after.
Found by a destructive-path matrix (25 scenarios, throwaway container, fake HOME
with sentinel files): main fails 10 (deletes unrelated dirs, files and symlink
targets at the plugin path); the series without this commit fails 2 (H8/H9);
with it, 25/25.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ordinary being