chore(memory): BDR-100 mechanisms, EVAL-030 self-audit, journal + TODO 2026-09-24
This commit is contained in:
@@ -1,5 +1,25 @@
|
||||
# TODO
|
||||
|
||||
## 2026-09-24 — root causes of the day's errors → mechanisms (feature/guardrail-evasion-citers)
|
||||
User: "détecte pourquoi tu as fait ces erreurs et corrige-les". Evidence: scratch
|
||||
`run-rc.sh` carries `GIT_CONFIG_GLOBAL=/dev/null` inline = the denied form my E2
|
||||
brief ordered ("exported first"); the 200-file rule and the density pass were
|
||||
patched from memory, never from a consumer grep. BDR-100, EVAL-030.
|
||||
- [x] R1 `settings.json` hard_deny "Routing around a guardrail" (wrapper/alias/
|
||||
heredoc/Makefile target/env file/other shell/other agent = same action;
|
||||
refusal → report + wait; brief ordering a refused form is wrong).
|
||||
- [x] R2 refusal clause in 14 agents (executors + reviewers) + CLAUDE.global.md
|
||||
sub-agent rule; hermetic tests only via `make test [suite=]`.
|
||||
- [x] R3 Makefile `make test suite=<file>` — the export lives in the Makefile.
|
||||
- [x] R4 `lib/tests/doctrine-citers.test.sh`: CLAUDE.md "Section" / § Label
|
||||
citations must resolve; flip-tested; first run fixed rest-api-node.md.
|
||||
- [x] R5 doctrine "After code changes" step 4: changed rule/heading/label/
|
||||
threshold → grep every citer, same commit; thresholds in one lib file.
|
||||
- [x] R6 verify: census 5/5, make test 168/170 (T16a pre-existing), suite= OK,
|
||||
shellcheck clean, CLAUDE.global.md 287 lines. UNMERGED — human gate.
|
||||
Residual: the content-aware PreToolUse guard (BLK-022) is still the missing
|
||||
deterministic floor for scripts run by a command; static deny stays string-based.
|
||||
|
||||
## 2026-09-24 — C2 coherence: 30 doctrine/skill tensions resolved (feature/c2-coherence)
|
||||
Audit by 3 read-only analysts (doctrine+rules, skills A-H, skills I-W), 39 raw
|
||||
pairs → 30 unique, spot-checked by grep. User approved all four groups + G3 as
|
||||
|
||||
Reference in New Issue
Block a user