With Auto-approve OFF the agent loop asked get_relevant_skills for
min_confidence=2.0, meant as 'published only'. The gate applies the
confidence check to published skills too, and confidence is 0-1, so no
user skill was ever injected, published or not. The explicit-activation
path already treats OFF as status == published.
Add a published_only flag to get_relevant_skills and pass it from both
callers that used the sentinel (relevance injection and the skill
preflight). Auto-approve ON is unchanged: a skill still needs a passing
audit and the confidence threshold.
Fixes#6634
The degenerate-stream guard treats a long run of one token, a dominant
token, or a periodic window as token collapse. Numbers repeat that way
by design: an 8x8 zero or identity matrix in LaTeX, a padded buffer, a
sparse 0/1 list. Each of those was stopped mid-answer with 'started
repeating tokens'.
Skip the same-run and dominance checks when the repeated token is a
number, skip the periodic check when the recent window is mostly
numbers, and skip the exact-block check when the block is all numbers.
A run of 2048 numbers in a row still stops, so a model stuck printing
'0 0 0' is still caught. Word loops and mixed word/number loops are
unchanged.
Fixes#5560
_connect_http entered streamablehttp_client and ClientSession on an
AsyncExitStack but never closed it when initialize or list_tools
failed. The half-open transport was left to the garbage collector,
which exited its AnyIO cancel scope from the wrong task, and the event
loop kept spinning: about 18% of a core per failed connect, growing
with each one, until restart.
Close the stack in the connecting task on any failure or cancellation,
the same way _connect_sse already does.
Fixes#5518
The exact-equality assert on the line above (requested_url == WEBHOOK_BASE)
already implies the URL has no trailing slash, so the endswith check adds
nothing.
_join_integration_url built urljoin(base + '/', '') for a bare '/'
path — the minimum execute_api_call accepts — so every request against
a POST-to-base integration went to base_url + '/'. Discord webhook
URLs 404 ('Unknown Webhook') on the trailing-slash variant, which made
the integration look broken even though the stored base URL was
correct.
Resolve a bare '/' (or empty) path to the base URL itself and keep all
other paths joining exactly as before, including deliberate trailing
slashes inside non-empty paths (linkding /api/tags/, Home Assistant
/api/). The reminder webhook sender and the discord_webhook
connectivity test already posted to the bare base URL; execute_api_call
was the remaining path that re-added the slash.
Fixes#5138
Harshit Ojha