263b32467d
feat(dashboard-agent): give the golden runs their own eval sample rate
...
The CI lane reads its own variable and defaults to every turn, so neither lane can change
the other's rate.
2026-08-06 00:11:25 +00:00
a4b4ef1b55
feat(webapp): let an organization turn off judging of its agent turns
...
A per-org feature flag the agent checks before every judged turn, and a turn is judged only
on an explicit yes.
2026-08-06 00:11:23 +00:00
3c5b2d5d6b
feat(dashboard-agent): never judge a turn that read source
2026-08-06 00:11:22 +00:00
42ddcffef0
feat(dashboard-agent): keep customer data out of the judged-turn row and the judge
...
Payloads, outputs, query rows and file contents are replaced by their shape before a turn
reaches the judge, and the row now keeps the derived verdict only.
2026-08-06 00:11:22 +00:00
7f13add1a0
feat(webapp): retire judged-turn rows after 30 days
...
The table is append-only quality data with no reader, so the sweep now drops rows past
the period in one bounded statement per run.
2026-08-06 00:11:21 +00:00
08dc7ba732
fix(webapp): keep the watch card in step with its draft, and frozen while it submits
2026-08-05 23:48:03 +00:00
c3ee6a4dbe
perf(webapp): resolve the agent card's links in one request
2026-08-05 23:48:02 +00:00
ebad471b43
fix(webapp): cap the size of a message to the agent
2026-08-05 23:48:01 +00:00
a0d06e4eec
fix(webapp): send the agent wake signal to a browser that has never opened the panel
2026-08-05 23:48:01 +00:00
18dffa6364
refactor(dashboard-agent): give the watch tests their own file
...
The wake narration and watch investigation cases move verbatim to
watch-actions.test.ts; the scaffolding both test files use moves to
test-support.ts. Same 146 passed / 2 skipped.
2026-08-05 23:42:35 +00:00
46290d1ca6
refactor(dashboard-agent): move the watch wake and investigate lanes out of the agent
...
Pure move. The wake narration and the consented investigation now live in
watch-actions.ts, reached from one call in onAction; the runtime they share with
the agent's own hooks moves to agent-runtime.ts. Every export is re-exported from
dashboard-agent.ts, so no import path changes.
2026-08-05 23:42:35 +00:00
2ed5d52260
refactor(dashboard-agent-db): move the watch tables into their own schema module
...
Pure move, re-exported from schema.ts so drizzle-kit reads the same six tables
and the generated SQL is unchanged.
2026-08-05 23:42:34 +00:00
30ec754a37
refactor(dashboard-agent-db): move the watch queries into their own module
...
Pure move: the watch, wake and batch-chain queries leave queries.ts for
watch-queries.ts, re-exported so every import path still resolves.
2026-08-05 23:42:33 +00:00
d6debe8fb7
test(webapp): cover the user-actor token's environment scope on every route that accepts one
2026-08-05 23:34:45 +00:00
5b5a25882f
fix(webapp): scope a delegated user-actor token to the environment it was minted for
2026-08-05 23:34:44 +00:00
b6ebe55500
fix(dashboard-agent): escape the prefix separator so the file reviews as text
2026-08-05 23:23:40 +00:00
a88228c838
chore(webapp): move the agent storybook gallery out to a follow-up
...
The gallery pages, their fixtures and the screenshot script only exist to
look at the agent UI, so they ship separately. The fixtures two unit tests
read — the report view models and the page contexts — stay here.
2026-08-05 22:30:45 +00:00
cafd920060
chore: keep the release notes as one
2026-08-05 22:18:03 +00:00
be1cb0458e
perf(dashboard-agent-db): index the open investigations the sweep scans
2026-08-05 22:17:43 +00:00
8b042bed30
perf(webapp): skip the admin read on the watch authorization path
2026-08-05 22:17:43 +00:00
1dd85367b7
perf(webapp): answer the wake poll with one narrow read per database
2026-08-05 22:17:42 +00:00
d26a28b0e6
perf(webapp): only poll the wake feed for a browser that knows a watch exists
2026-08-05 22:17:41 +00:00
e109a243f5
fix(webapp): read the alert channel and the watch target on the primary
2026-08-05 22:17:41 +00:00
fc4381d50e
fix(webapp): keep the agent wake poll to one chain per tab
2026-08-05 22:17:40 +00:00
9b4e4734a1
perf(dashboard-agent): share one 1h-TTL prompt cache between head start and the agent
...
Head start passed `system` as a bare string, so Anthropic neither wrote nor read
the cache: the ~18.4k-token prefix was billed in full and the agent's step 2 then
paid for a fresh write. Both breakpoints now carry a 1-hour TTL.
The two prefixes were also not identical — `get_run` sat in a different position in
the agent's tool set than in the canonical schema-only one, so they could never have
shared a cache. Reordered, and a fingerprint over system text + tool definitions is
logged per model call alongside the provider's cache-write, cache-read and uncached
input token counts, so a future drift is visible instead of silent.
2026-08-05 22:16:30 +00:00
893b9e866c
chore(dashboard-agent): oxfmt
2026-08-05 22:16:29 +00:00
e8d35fab8c
perf(dashboard-agent): cap read_file at 48KB / 1500 lines
...
A single read could put ~65k tokens in the transcript, paid for on every later
turn. The truncation notice points at the line range, and the range is now applied
before the cap so a read past the ceiling returns the lines asked for.
2026-08-05 22:16:29 +00:00
7850289107
perf(dashboard-agent): stop echoing the panel's view model back to the model
...
render_view and get_report both return the panel's view model. Their toModelOutput
now gives the model an acknowledgement and the graded summary respectively, while
the client still receives the blocks and the full report. Errors pass through
verbatim, so the fix-and-render-again loop is unchanged.
2026-08-05 22:16:28 +00:00
dd1f309bae
perf(dashboard-agent): cap the judge's tool payload and stop pretty-printing it
...
Each tool output is capped at 1500 characters with a truncation marker, and the
prompt no longer indents the whole turn of tool traffic.
2026-08-05 22:16:27 +00:00
02debc62d9
perf(dashboard-agent): sample a tenth of turns for the judge by default
...
The per-turn LLM judge was billing every turn, and an unparseable
DASHBOARD_AGENT_EVAL_SAMPLE_RATE fell back to full rate too. Nothing reads
chat_turn_evals yet, so 10% loses no signal we consume.
2026-08-05 22:16:27 +00:00
6a6fa1f4fd
chore: keep the release notes as one
2026-08-05 22:13:07 +00:00
d4f0357ac1
chore(webapp): move the gallery's transcripts into the storybook fixtures and drop demo-chats
2026-08-05 22:12:44 +00:00
a7a29f368f
chore: add a server-changes note for the report card fixes
2026-08-05 22:12:43 +00:00
0760f1b55d
refactor(webapp): give the agent cards one card primitive
2026-08-05 22:12:42 +00:00
3f85acc2a7
fix(webapp): give the diagnosis card links a focus ring via a TextLink variant
2026-08-05 22:12:41 +00:00
5a80876b16
fix(webapp): ring the unread dot against the navbar surface
2026-08-05 22:12:40 +00:00
9b32bfda66
fix(webapp): let agent chips follow the interface-contrast control
2026-08-05 22:12:39 +00:00
fd83288906
fix(webapp): drop the metric sparkline to its own line on a narrow panel
2026-08-05 22:12:38 +00:00
3773f31d22
fix(webapp): route the report's stale-data badge through AgentBadge
2026-08-05 22:12:37 +00:00
5800e8929d
chore: drop the unrendered demo cards and fold the release notes back into one
2026-08-05 21:49:37 +00:00
8d3676a874
chore: oxfmt the lines the removed comments let collapse
2026-08-05 21:38:32 +00:00
f9eb9c0896
chore(dashboard-agent-db,contracts): cut the query and wire-contract comments
2026-08-05 21:38:25 +00:00
9270caf715
chore(webapp): cut the watch service and route comments to their invariants
2026-08-05 21:38:24 +00:00
42749c9772
chore(dashboard-agent): cut the agent, tool and watch-tick comments to their invariants
2026-08-05 21:38:23 +00:00
5153cca045
chore: restore the comments an off-by-one over-trimmed
2026-08-05 21:38:13 +00:00
9b7ca8c1de
chore: bring the last three-line comments down to two
2026-08-05 21:37:46 +00:00
e6ec1d2d11
chore(webapp): delete the agent panel and view rationale comments
2026-08-05 21:37:45 +00:00
a4f8ed1906
chore(webapp): trim the seed-script narration
2026-08-05 21:37:09 +00:00
0739337c24
chore: cut the report, waiting-run and package comments to their facts
2026-08-05 21:37:08 +00:00
93bc6c158d
chore(webapp): clear the demo fixture and storybook commentary
2026-08-05 21:36:19 +00:00
Katia Bulatova