Learn Template Session Fixes

Seven template text fixes based on measured failures across 3 Learn sessions and 42 pipeline runs. Each change addresses a specific behavior that failed in practice — not theoretical improvements. Zero CLI code.

verdict PASSscore 13 / 13findings 3 (0 risk · 1 debt · 2 obs)duration 3h 1mrejection cycles 0shipped May 2, 2026surface cli

Pipeline timeline

Intent to proven code in 3h 1m across Think, Plan, Build, and Verify.

Think
164m
Plan
164m
Build
7m
Verify
9m

Assertion ledger

13 claims, each independently verified. Showing 8 — show all →

IDSaysMatcher
A001Accept-action intro explains that accept does not mean closeverifiedok
A002Accept-action intro instructs re-evaluation on own meritsverifiedok
A003First/routine session calibration lines are removedverifiedok
A004Garden size calibrations still exist after removing first/routineverifiedok
A005Observation prompt exists as a standalone instruction before the format blockverifiedok
A006Session approach uses priority ordering instead of a flat capverifiedok
A007The arbitrary 30-finding cap is removedverifiedok
A008Session wrap-up offers a Think handoff as an explicit optionverifiedok

Findings 3 total

debtpackages/cli/templates/.claude/agents/ana-learn.md→ closed
Edge Cases section still says 'Cap at ~30 per session' — contradicts new 'no arbitrary cap' guidance in Session Approach
obs→ closed
Spec Change 4 only targeted the Session Approach paragraph but didn't address the matching ~30 reference in Edge Cases — scope gap
obs→ closed
Proof context finding about SKILL.md path naming still present — not addressed by this build (expected, out of scope)

Integrity seal

scopesha256:b278eeee9850b...
contractsha256:4de2217743f2d...
plansha256:3a578b3d34569...
specsha256:d729fad1441eb...
build-reportsha256:6727013be5dd0...
build-datasha256:5bdcbf1cfc119...
verify-reportsha256:28298ba7a1b37...
verify-datasha256:6d672799b488d...
audit cmd$ ana proof audit learn-template-session-fixes   → all hashes match