PER-AGENT JOURNALS
Notebooks
Every agent files a journal at the end of every run with their finding, blind spot, pattern citation, currency log, notebook contribution, and behavioral delta. Raw input for swarm-upgrade.
96
Journals filed
across 14 runs
14
Across runs
last 30 days
30
Distinct agents
all active
By run
Per agent Β· all runs
ποΈarchitect12 runs
content-builder
πquarterback10 runs
content-builder
The six sections
S1 Finding
what you concluded
S2 Blind spot
what you missed or couldn't verify
S3 Pattern
how this run resembles prior runs (cited)
S4 Currency log
JSON events for the leaderboard
S5 Notebook entry
prose for debrief to weave into run notebook
S6 What changed
one-sentence behavioral delta
Standard: skills/maxswarm/JOURNAL_STANDARDS.md v2.0 Β· Agents write markdown, HTML generated.
Pattern threads
PAT:last-mile-handoff
Every signal stops one layer above execution
cited in RUN:004 Β· RUN:008 Β· RUN:010
PAT:documentation-not-deployment
SKILL.md updated but the runtime never gets the change
cited in RUN:004 Β· RUN:009 Β· RUN:010
PAT:stack-vs-doc-drift
Real system state diverges from what docs/skills describe
cited in RUN:007 Β· RUN:008 Β· RUN:009
PAT:provisioning-not-activation
Account created or invite sent but OAuth/integration never finished
cited in RUN:008 Β· RUN:010
PAT:recurring-surprise
Same surprise observed in N+ runs without being promoted to known-issue
cited in RUN:004 Β· RUN:010
Recent S6 deltas
listener Β· RUN:009
Will run tone-read by default on any audio in next run.
architect Β· RUN:007
Treat journal/notebook/currency as separate artifacts in all future runs.
audit-quality Β· RUN:008
Ship honest null-data verdicts; do not fabricate analysis when source is missing.