A new package, nisse-ui: a read-only page for studying how the agents
work. For each work item it shows the steps (what people and the
orchestrator said, and each run), each run's turns, and each turn's
request read against the log it was made from:
- each message beside the events it came from, tagged unchanged,
labelled, narrated, shortened (by how much) or a compaction, with what
fitting dropped;
- what is new since the turn before, and where the request first
differs (where prompt reuse had to stop), next to the server's count;
- the reply with its reasoning, and the raw JSON.
Runs going in circles are flagged by nisse_core.loop.circles (repeated
near-identical replies, one correction given again and again, repeated
apologies), a pure function M7.8 can reuse in Driver.run.
In nisse-core:
- render.py: message_indexes(), from the same code render() uses, so
the page's matching can't drift from what was sent.
- calls.py: the record notes when recording began; the reader gains
calls_for(item) and call(id).
- state/readonly.py: opens an agent's SQLite file without creating its
-wal/-shm files (mode=ro alone does, once the writer has closed).
The page reads only through read-only connections, escapes everything,
and sends a Content Security Policy that allows no script. It listens on
127.0.0.1; until M6.6d puts it behind Authelia, reach it over an SSH
tunnel. Jinja2 is its one new dependency, for autoescaping.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J5Utyfzw7gF7UhjRDEtjTk
nisse-core records each model call of every agent into a file beside its
state store (<state>.calls.db), for the status page (L7, X6):
- the request body exactly as sent (never headers), the raw response with
its reasoning, the model asked and the one that answered, time, tokens
and llama-server's own prompt reuse (timings.cache_n);
- its work item, role, run and purpose: Driver.run and nisse-dev's
delegation loop open runs, nested; compaction, expert consults, the
classifier and the go-ahead reader label their calls;
- provenance, worked out in BudgetedClient by comparing the events at
each stage: each event sent with its log id and content as sent, which
a compaction replaced, which fitting shortened (by how much) and
dropped.
Content is stored once by hash, compressed. It is diagnostic only: the
agent never reads it, a failed write is logged once and recording stops
while the call goes on, and the file is deleted when the agent starts,
so its layout never migrates; CallReader refuses a layout it doesn't
know. While recording, the system prompt ends with one sentence saying
so. `call_record: false` in either agent's YAML turns it off.
The benchmark records each run beside that run's own state store, and
`--keep-runs DIR` keeps both for the page.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J5Utyfzw7gF7UhjRDEtjTk
dongarandClaude Opus 5.5