fix: repair runtime skill boundaries and record verified closure · TeaEntityLab/teaPrompt@f86173b · GitHub
Skip to content

Commit f86173b

Browse files
johnteeejohnteee
committed
fix: repair runtime skill boundaries and record verified closure
Validate canonical execution caps before arithmetic and dispatch, preserve backlog queue errors and completion ledgers, and distinguish worker failures from configuration holds. Separate private scoring order from public arm order, enforce blinded scorer-data paths, publish a runnable CONFIG, and normalize router confidence/review inputs with action-scoped hazard checks. Apply the explicitly approved five-to-ten domain-pack oracle migration while preserving nine core skills, locked: true, and the restored read-only oracle mode. Record all finding dispositions, decision rationale, failure provenance, verification limits, and separate commit/push authorization. Refresh live verification maps and regenerate discovery metadata after the final document edits. Verification: final-state generate_index.py + make all: 1533 passed, zero validator errors, 19 valid contracts, examples for 9 core + 10 packs, and 100% ROUTE-001/002/003 fixture consistency. Markdown consumer and Chromium verified both new record sections and four two-column advisory rows with no page overflow; a clean viewport capture was retained. Eight length warnings and 35 historical hygiene warnings remain visible. No provider efficacy, host isolation, statistical blinding, or external-effect safety claim.
1 parent 13fc95c commit f86173b

25 files changed

Lines changed: 1545 additions & 225 deletions

‎VERIFY.md‎

Lines changed: 6 additions & 5 deletions

‎acceptance.yaml‎

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,8 @@
11
# acceptance.yaml — acceptance spec for TeaPrompt (read-only by convention)
22
# Generated by verification-map-generator, 2026-09-23; route-evals check widened
33
# to all three fixtures 2026-09-24 (source_commit 7d8de92).
4+
# Domain-pack oracle migration approved by the user on 2026-10-07:
5+
# five -> ten admitted packs; locked:true and all other checks remain unchanged.
46
# Oracle rule: a spec-vs-repo mismatch is a spec-oracle error or a product
57
# regression — never edit this file to make a run pass. Git does not preserve
68
# file modes, so the read-only status is a convention, not a mechanism; edits
@@ -12,9 +14,9 @@ checks:
1214
verify: python3 -c "import re; s=open('reflective-prompt-library/plans/validate_skill_examples.py').read(); print(len(re.findall(r'\"([a-z-]+)\"', s.split('CORE_SKILLS = [')[1].split(']')[0])))"
1315
expect: "9"
1416
- id: domain-pack-cardinality
15-
description: DOMAIN_PACK_SKILLS lists exactly 5 packs
17+
description: DOMAIN_PACK_SKILLS lists exactly 10 packs
1618
verify: python3 -c "import re; s=open('reflective-prompt-library/plans/validate_skill_examples.py').read(); print(len(re.findall(r'\"([a-z-]+)\"', s.split('DOMAIN_PACK_SKILLS = [')[1].split(']')[0])))"
17-
expect: "5"
19+
expect: "10"
1820
- id: gate-suite-green
1921
description: make all exits 0
2022
verify: make all

‎features/registry.md‎

Lines changed: 13 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1,35 +1,41 @@
11
---
22
feature: skill registry cardinality
3-
source_commit: 7d8de92
4-
last_verified_at: 2026-09-24
3+
source_commit: 13fc95c
4+
last_verified_at: 2026-10-07
55
verification_status: passing
66
---
77

88
# Skill registry cardinality
99

10+
Human-reviewed RV-05 migration approved 2026-10-07: ten admitted domain packs,
11+
nine core skills, and nineteen installed skill contracts. `source_commit` names
12+
the pre-repair base; this migration is currently a working-tree change.
13+
1014
## Entry points
1115

1216
- `reflective-prompt-library/plans/validate_skill_examples.py` — `CORE_SKILLS`
13-
(frozen nine) and `DOMAIN_PACK_SKILLS` (five) lists.
17+
(frozen nine) and `DOMAIN_PACK_SKILLS` (ten) lists.
1418
- `reflective-prompt-library/skills/` — directories must match the registry.
1519

1620
## Drive
1721

1822
```bash
19-
find reflective-prompt-library/skills -mindepth 2 -maxdepth 2 -name SKILL.md | wc -l # expect: 14 (9 core + 5 packs)
23+
find reflective-prompt-library/skills -mindepth 2 -maxdepth 2 -name SKILL.md | wc -l # expect: 19 (9 core + 10 packs)
2024
python3 reflective-prompt-library/plans/validate_skill_examples.py
2125
```
2226

23-
Expected validator tail: `All 9 core + 5 domain-pack skills have example files`.
27+
Expected validator tail: `All 9 core + 10 domain-pack skills have example files`.
2428

2529
## Observable outcomes
2630

2731
- Exactly 9 core skills: reflective-dispatch, reflective-brief,
2832
reflective-spec-plan, reflective-implement, reflective-minimality,
2933
reflective-review, reflective-research, reflective-risk,
3034
reflective-handoff-retro.
31-
- Exactly 5 domain packs: flow-control-generator, flow-loop-harness,
32-
agent-governance-scaffold, governed-delivery, verification-map-generator.
35+
- Exactly 10 domain packs: flow-control-generator, flow-loop-harness,
36+
agent-governance-scaffold, governed-delivery, verification-map-generator,
37+
headless-agent-cli-contract, arm-blinded-eval-harness,
38+
acceptance-join-validator, golden-benchmark-runner, router-trace-linter.
3339
- Every registered skill has `skills/<name>/SKILL.md` and
3440
`skills/examples/<name>.examples.md`.
3541

‎features/test-suite.md‎

Lines changed: 13 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -1,31 +1,33 @@
11
---
22
feature: governance test suite
3-
source_commit: f0f986d98af86702c5ecacac6daf095a06849686
4-
last_verified_at: 2026-10-02
3+
source_commit: 13fc95c
4+
last_verified_at: 2026-10-07
55
verification_status: passing
66
---
77

88
# Governance test suite
99

1010
## Entry points
1111

12-
- `reflective-prompt-library/plans/tests/` — 106 `test_*.py` files; the
13-
2026-10-02 verification drive reported 1358 passing tests. Pins governance
14-
records, registry cardinality, cheatsheet parity, boundary quick cues and
15-
adoption-state contracts.
12+
- `reflective-prompt-library/plans/tests/` — governance records, registry
13+
cardinality, cheatsheet parity, boundary quick cues, adoption-state
14+
contracts, and emitted-script consumer regressions.
1615

17-
Generation-time baseline (2026-09-24, `7147e91`): 99 files and 1290 tests.
18-
The metadata now names the current committed source baseline; this repair
19-
changes covered documentation, not test code. Run receipts and limits are
20-
recorded in `review/final-report.md`.
16+
Historical snapshots: the generation-time baseline (2026-09-24, `7147e91`)
17+
had 99 files and 1290 tests; the 2026-10-02 drive reported 106 files and
18+
1358 passing tests.
19+
`source_commit` now names the committed base; the verified working tree
20+
includes the current runtime-review and consumer-regression repairs.
21+
This map refresh changes documentation, not test code. Run receipts and
22+
limits are recorded in `review/final-report.md`.
2123

2224
## Drive
2325

2426
```bash
2527
python3 -m pytest reflective-prompt-library/plans/tests/ -q
2628
```
2729

28-
Observed tail on 2026-10-02: `1358 passed`. The invariant is **0 failures**,
30+
Observed tail on 2026-10-07: `1533 passed`. The invariant is **0 failures**,
2931
not the exact count; re-verify after covered source changes.
3032

3133
Per-area selection uses `-k` or a file path, e.g.:

‎features/validators.md‎

Lines changed: 12 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,12 +1,16 @@
11
---
22
feature: standalone validators
3-
source_commit: 7147e91
4-
last_verified_at: 2026-09-24
3+
source_commit: 13fc95c
4+
last_verified_at: 2026-10-07
55
verification_status: passing
66
---
77

88
# Standalone validators
99

10+
`source_commit` names the committed base. The verification includes the
11+
current uncommitted runtime-review repairs; this map refresh changes
12+
documentation, not validator behavior or acceptance rules.
13+
1014
## Entry points
1115

1216
Eight scripts in `reflective-prompt-library/plans/`, each runnable solo:
@@ -26,9 +30,12 @@ python3 $P/validate_route_fixture.py # ROUTE-002/003 minimum counts
2630
## Observable outcomes
2731

2832
- Every script exits 0 and prints a ✅-style success line.
29-
- `lint_skills.py` exits 0 with **one known warning**
30-
(`agent-governance-scaffold` body length, accepted R10 debt). A *second*
31-
warning or any error is a finding.
33+
- `lint_skills.py` exits 0 when it reports warnings but no errors. The
34+
2026-10-07 drive reported eight long-skill warnings across six packs;
35+
the full list is retained in the runtime-review handoff.
36+
This observation is not an acceptance waiver: R10 covers only
37+
`agent-governance-scaffold` body length, and new or changed warnings remain
38+
findings to triage. Any error is a failure.
3239

3340
## Failure paths
3441

‎reflective-prompt-library/PROJECT_KNOWLEDGE.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -134,7 +134,7 @@ deferred promotions are recurrence-gated — see [panel backlog](plans/multi-age
134134
> Pointers to the causal trail — plans, reflections, tests, commits. Detail is
135135
> not duplicated here; this is a map, not an archive. Mechanical panel rounds
136136
> are rolled up; the round-by-round trail lives in the panel record.
137-
- 2026-10-07 Recent-change review and recording handoff — eight commits `c454738`–`2989110` plus the initial five survey/navigation files reviewed; archived consumer probes found numeric-cap, scoring-order/input-boundary, router R4, queue-read, and locked-oracle defects. Survey facts/table rendering corrected during recording; runtime fixes and five-to-ten locked-oracle migration remain unimplemented. Historical green gates are not closure evidence for those cases → [handoff](plans/recent-changes-review-handoff-2026-10-07.md)
137+
- 2026-10-07 Recent-change review and recording handoff — eight commits `c454738`–`2989110` plus the initial five survey/navigation files reviewed; archived consumer probes found numeric-cap, scoring-order/input-boundary, router R4, queue-read, and locked-oracle defects. Survey facts/table rendering were corrected during recording; subsequent user-directed repairs closed all ten runtime findings and the explicitly approved five-to-ten oracle migration with fresh generated-script/consumer evidence. Historical green gates are not closure evidence; host isolation, comparative efficacy, and provider utility remain untested → [handoff](plans/recent-changes-review-handoff-2026-10-07.md)
138138
- 2026-10-07 Diagram Design survey — `cathrynlavery/diagram-design` @ `d1376371`, manifest 2.6.64; 44 visual types verified by reference inventory and independent guide extraction; 15/15 bounded helper probes and real Chromium playback/static/fallback/export checks. Semantic-first selection, fidelity ledgers and layered verification retained as references; mobile overflow and Google Fonts dependence recorded; no skill/pack/runtime adoption or efficacy claim → [record](plans/diagram-design-survey-2026-10-07.md)
139139
- 2026-10-07 Cross-survey rethink on deferred candidates — 81 ledger-bearing documents re-examined against installed surface post-five-pack admission; RTH-1 adopted in place under user direction: eval measurement-discipline repair on golden-benchmark-runner and arm-blinded-eval-harness (noise floor basis, failure categorization, selection-vs-final separation); RTH-2 (tenth core skill) and RTH-3 (RRSI annealed budget) rejected; core remains frozen at nine → [record](plans/cross-survey-rethink-2026-10-07.md)
140140
- 2026-10-07 System prompts leaks survey — `asgeirtj/system_prompts_leaks` @ `f015440d`; 1,237 file-like entries within 16 directory buckets, not a vendor census or authenticated prompt count. Four-lens source review found progressive-disclosure and authority-boundary patterns; vendor enforcement, extraction success and comparative efficacy remain unverified. SPL-1 reference-only, zero operational prompt text adopted → [record](plans/system-prompts-leaks-survey-2026-10-07.md)

‎reflective-prompt-library/index.json‎

Lines changed: 56 additions & 27 deletions

0 commit comments

Comments
 (0)