BRIEF-vurdering-v2.md tiltak 2: "design for the investigator, not the
validator". The hook alerts stated short conclusions a reader can only
accept or dismiss ("Rapid-fire: N consecutive fast interactions",
"possible stuck/spiral", "Consider a break").
Each alert in tool-tracker.mjs now follows the form the read-dominant
edit-ratio message introduced in 2c9e2de — observation, the counter-signal
that changes how to read it, then what to check:
- burst: names the interval and that edits were among the calls, and asks
whether each change was verified before the next
- edit ratio: carries the read percentage and asks what the remaining
calls are doing and whether the approach is converging
- soft warning: closes on framing instead of prescribing a break
- hard warning: asks the model to name its observations and ask what they
reflect; the required stop action is unchanged
commands/interaction-report.md gains the same rule for Observations and
trend reporting, plus an explicit "investigator, not validator" tone rule
and a "report the difference, not a label for it" rule.
Wording only — thresholds, heuristics, data model and required actions
are untouched. README examples and threshold-basis cells updated to match.
Tests first (Iron Law): 6 new/updated assertions on message text in
tests/tool-tracker.test.mjs and tests/interaction-report.test.mjs, red
before the change. node --test tests/*.test.mjs: 269 pass, 5 fail — the
pre-existing perf wall-clock cases only.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013U8ZH25KiMtts89yWRuVWD
The burst counter incremented on any tool call <30s apart regardless of
tool type, so a bulk read of many files produced the same "Rapid-fire"
alert as a rapid-fire editing sequence. The edit-ratio check reported
"possible stuck/spiral" for any session under 10% edits past 30 minutes,
which a read-heavy analysis session satisfies structurally whether or not
the work is productive. Both mechanisms are confirmed in
docs/BRIEF-vurdering-v2.md, verification points 1a and 1b.
tool-tracker.mjs now tracks whether the current burst run consists only
of read tools and suppresses the rapid-fire alert for such runs. A
read-dominant session (>=70% reads) reports its low edit ratio as
context rather than as a stuck/spiral claim, and never as the sole
reason for a warning. Bursts involving Edit/Write/Bash and
non-read-dominant sessions behave exactly as before.
No new data is recorded: the differentiation uses tool_name, which
events.jsonl already logs. The privacy design is untouched.
Tests first (Iron Law): 5 cases in tests/tool-tracker.test.mjs, 3 of
which failed against the previous implementation, 2 of which are
regression guards that passed before and still pass.
Verified: node --test tests/*.test.mjs -> 263 pass, 5 fail. All 5
failures are the perf wall-clock assertions, which fail identically on
HEAD without this change (measured by stashing and re-running) — they
are a pre-existing property of this machine, not a regression.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011wMhJuPFUiWqKLSQjrzCXP
Widens the F-3 sweep from the reference list to every research claim in
README.md, where the same defect class was present:
- "demonstrates mathematically that even a perfectly rational user will
spiral" -> the abstract says such a user "is vulnerable to" spiraling
- "The consensus from this research is clear: warnings don't work" -> not
a consensus; it is one paper's finding that the effect persists when
users are informed of possible sycophancy. Re-attributed.
- "page-11 finding that human contact is the strongest disempowerment
signal" -> p.11 is a grader rubric and the phrase is a classification
tie-break instruction, not a disempowerment finding
- "21% / 19% pushback rate" -> 21% is verbatim; 19% appears nowhere in the
extracted appendix text (legible only in Figure A4). Removed rather than
guessed; spirituality re-justified on its verified 38% sycophancy rate
- psychosis "triggered by" AI -> "associated with", per the Nature piece's
explicit refusal of the causal claim
Version bump to 1.2.2 (plugin.json, README badge, CHANGELOG). SKILL.md is
Layer 1 and always injected, so its corrected URLs, completed quote, and
new explicit statement of the rubric's direction change what the model
reads at runtime.
Tests: 257/258 (the 1 red is the known perf wall-clock flake).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011rtdS8Ufpen419n6R9HyMm
Bump plugin.json + README badge to 1.2.1 and cut CHANGELOG [1.2.1].
The actual fix (hookEventName in hookSpecificOutput) landed in 7d53a73;
this release ships it so the catalog ref can move off the buggy v1.2.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DjqAo7X5tj36gtHXtP3RoY
outputWithContext() emitted a hookSpecificOutput object without the
required hookEventName field, so Claude Code rejected the hook output
("hookSpecificOutput is missing required field hookEventName") and
dropped the context the SessionStart, UserPromptSubmit, and PostToolUse
hooks tried to inject. Thread the event name through the helper; each
caller now passes its own (SessionStart / UserPromptSubmit / PostToolUse).
Existing emission tests now assert hookEventName per event.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VmHCQjJHUyWwxGAVVjNLgp
Add /ultraresearch-local for structured research combining local codebase
analysis with external knowledge via parallel agent swarms. Produces research
briefs with triangulation, confidence ratings, and source quality assessment.
New command: /ultraresearch-local with modes --quick, --local, --external, --fg.
New agents: research-orchestrator (opus), docs-researcher, community-researcher,
security-researcher, contrarian-researcher, gemini-bridge (all sonnet).
New template: research-brief-template.md.
Integration: --research flag in /ultraplan-local accepts pre-built research
briefs (up to 3), enriches the interview and exploration phases. Planning
orchestrator cross-references brief findings during synthesis.
Design principle: Context Engineering — right information to right agent at
right time. Research briefs are structured artifacts in the pipeline:
ultraresearch → brief → ultraplan --research → plan → ultraexecute.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
