Skip to content

#381 pre-run record: Amendment 3 merge checks and recovered inputs; #462 uninstall note #466

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
seathatflowsinourveins merged 2 commits into from
Sep 28, 2026
Merged

#381 pre-run record: Amendment 3 merge checks and recovered inputs; #462 uninstall note #466

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cd5f68c
#381 pre-run record: Amendment 3 merge checks, recovered command iden…
Sep 28, 2026
208ffb5
Evidence manifest: re-register the #381 README and the delegated-deci…
Sep 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions docs/decisions/2026-09-28-delegated-decisions.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,10 @@ The returned lane output is in
- the prose at `examples/claude-native/workflows/README.md:19,81`, `recipes/claude-native-ultracode.md:156`, `adoption/bootstrap.md:346-347`, `catalogs/foundation/practice-references.json:73` and `docs/community-native-practice.md:190`;
- an evidence sentence added to `adoption/hooks/claude/token-lanes-block.builder.md`. #456 left it out only because it duplicated the preload.
The host uninstall is the upstream `skills remove verification-before-completion -g -y`, run after merge, because `install_skills.py` has no remove mode.
**Dated note, 2026-09-28 (after #464 merged at `c0966da2`):**
- This unscoped form is the complete one. Its one precondition is to assert first that no other agent holds a same-named copy, because without `-a` skills 1.7.0 targets every agent (`dist/cli.mjs` L6834-6838).
- An agent-scoped `-a claude-code codex` does not remove the skill. A remaining detected universal agent resolves to the canonical `.agents/skills` directory (`getAgentBaseDir` L2214, `isUniversalAgent` L2180). That sets `isStillUsed`, which keeps the canonical folder and its lock entry (L6883-6907).
- The skills-trial owner found this during the uninstall. It was re-read from the `skills@1.7.0` npm package.
- **#381 Amendment 3.** The #381 role-body table pins `isolated-builder.md` at `57452a64…`, and main still equals it. The removal therefore lands with a dated pre-execution Amendment 3 in the same PR, authorised by the Gate A owner.

### #444: acknowledgement from the live trading lane
Expand Down
104 changes: 104 additions & 0 deletions evidence/artifacts/token-adoption-e2e-20260926/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -773,3 +773,107 @@ Amendment 2 table for the launch check; the **Sealing** rules apply unchanged.

Replacement hash: `RUNBOOK.md` `33810ef7…` → `135340b6…`.
`preregistration.json`, `token-e2e-run.mjs` and both fixtures are unchanged.

## Pre-run record (2026-09-28): Amendment 3 merge and recovered inputs

This dated record is not an amendment. No task text, information need, check,
arm or route changes, and the five sealed files are untouched. No capability
probe or organic arm has run.

**Amendment 3 merge.** #464 merged Amendment 3 at `c0966da2`
(2026-09-28T18:29:54Z). That revision is the `preregistration_commit` checked
at launch. The following were re-verified at `c0966da2`:
- The reused table pointer serializes to the sealed `frozen_input`:
`6899e551b2bea3918b47cba1b41e27e59d8f5cb12a681f98cc77f73d66a2a2c1`, 6,552
bytes, 60 records.
- All three copies of each of the five role bodies equal their pinned rows:
Amendment 2's table for four of them, and Amendment 3's row for
`isolated-builder.md`.
- `python3 -B -m unittest tests.test_token_e2e_preregistration` ran 34 tests,
OK.

The six token-lanes blocks at that revision, for the launch's carrier check:

| Block | SHA256 | Bytes |
| --- | --- | --- |
| `token-lanes-block.md` | `d6c3c19d6e94d3f48d2788963c8c9be45ec57e6bb81cab92b2c30d9fd5c8e8d6` | 4,088 |
| `token-lanes-block.builder.md` | `c81a91c4ffb2fa906feb403d5a11bd294b6cf2db45795d80f9acd662c5004973` | 2,877 |
| `token-lanes-block.researcher.md` | `47cc80b262468557b8cfe5c243598030757270674bc7bf286fba8c0c573abc28` | 3,075 |
| `token-lanes-block.reviewer.md` | `19857a44a201948aa1d1fc9be3e24d5f3f7cc385f333ab3893411889258af4ed` | 2,088 |
| `token-lanes-block.scout.md` | `81784c2edc0cc95c0c0eb3b443b5b228dd41e970236239553c66af90ddfc7eae` | 1,089 |
| `token-lanes-block.verifier.md` | `a2eb3298126efd82622150e5717f10eb700f4aa3076d027c86ae1fe653f29eb0` | 2,186 |

On the workstation host, the installed copies of these six blocks and of the
builder body equal these bytes (read back on 2026-09-28). Another host
establishes its own read-back.

**Pre-run inputs.** Two pre-run input gates are satisfied here:
- the command-fidelity gate for `reuse-296-00` and `reuse-343-13`, whose
`pre_run_input_gate` reads “Recover and seal the exact six original
observation identities and inputs before capability probes”;
- the retained history report: “A missing retained history report also
blocks launch”.

**Command identities (`reuse-296-00`, `reuse-343-13`).** Both source runs'
original work files survived in an earlier session's scratch directory. Their
SHA256 values equal the receipts' `exact_comparison` digests:

| Source | Baseline (raw) | Filtered output |
| --- | --- | --- |
| #296 `/tools/0`, run at `f5812d3f` | `765a3f2e5dc893fb0a95413d583055de47319140a8c760a07ada6c300442ecf5` | `b8b26c4238e3e8d9d3fbfeba8fe810357ea1269b27597bcc86e98bde8b2ecb07` |
| #343 `/tools/13`, run at `c09dd6df` | `9c2e4e94e8b718541271823a9f6ec42653ed2f742d9d1c688eab3700ab73f330` | `4e8cb518f78cb024e28f9ed6e31eb832e3dbf33622cb171d441ed995df00b1c4` |

The six identities, in their original order, and the information need each one
serves in the frozen task text:

| # | Command | Information need |
| --- | --- | --- |
| 1 | `git log -30` | newest commit subject |
| 2 | `git status` | branch state |
| 3 | `git diff HEAD~5 --stat` | changed-file summary over the last five revisions |
| 4 | `grep -rn "def register_file" scripts` | matches for `def register_file` under `scripts` |
| 5 | `ls -la scripts` | directory inventory |
| 6 | `python3 -m unittest tests.test_host_requests` | test count and result status |

#296's digest-verified baseline carries these six commands as its section
headers, `=== [1] git log -30 ===` through
`=== [6] python3 -m unittest tests.test_host_requests ===`. #343's run result
lists 13 commands, and the receipt's `commands` field kept only the first
eight (with the scratch path sanitized), which end at command 4. Its
digest-verified baseline has no headers, but it contains the `ls -la` listing (a `total` line and directory entries) and the
unittest result lines `Ran 49 tests` and `OK (skipped=1)`. Commands 5 and 6 are
therefore covered by hashed bytes as well.

The identity is the bare command; the channel is provenance. #296 ran each
command as plain Bash inside an Ultracode subagent: the RTK hook rewrote it, and
the baseline came from `rtk proxy`. #343 ran `timeout 60s rtk <command>`
(`timeout 180s` for the unittest) with an environment prefix and `</dev/null`,
and its baseline came from `rtk proxy`. The historical counts (49 tests,
`OK (skipped=1)`) are not the new run's expected values. The checker compares
each arm's answer with that run's own retained raw originals.

**History originals (`reuse-296-02`, `reuse-343-06`).** Each receipt's
`baseline_sha256` is reproduced exactly by
`git --no-pager log --stat -150 <revision>` at that receipt's own
`catalog_revision`. The match was found by hashing this output for all 597
commits on all refs in this checkout, with committer dates from 2026-09-24T00:00 to
2026-09-27T12:00 (host local time), with git 2.43.0:

| Task | Source entry | Revision | SHA256 | Bytes |
| --- | --- | --- | --- | --- |
| `reuse-296-02` | #296 `/tools/2` | `f5812d3f2266` | `82e9249222bafd5daee41ee74840a92d2a00feabd70f4e1149adb21a600d630b` | 891,615 |
| `reuse-343-06` | #343 `/tools/6` | `c09dd6dfc68d` | `5939b5451e55a3744b7a8f299733904979046974a5c2aea340ecb6b3b9149c2c` | 774,521 |

The reports are retained privately and not committed, because they carry commit
author identities. The byte match holds under this host's git version and
configuration. After a git upgrade, or on another host, verify the retained
bytes against these digests instead of regenerating them. Each launch maps
the tasks to the retained files in `input_paths` (RUNBOOK run-configuration
table: history originals keep identical bytes and hash across arms). Verify
those files against these digests before probes.

**Still open before launch:**
- PR-A's residual measurement scope (#432's Residuals) and its regenerated
sanitized baseline, both merge-before-run;
- the Codex half of the collector join;
- a sealed quiet window W.
8 changes: 4 additions & 4 deletions manifests/evidence.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14050,8 +14050,8 @@
},
{
"path": "docs/decisions/2026-09-28-delegated-decisions.md",
"sha256": "984cd31892637502ec50e2499c15485063beb78e44e397bb12860d489c61c50b",
"bytes": 16000
"sha256": "ef93948f54a2ed1695a1d332f01423d2ba6266c5429146fe547142c16e986698",
"bytes": 16692
},
{
"path": "docs/decisions/2026-09-28-top-rule-templates.md",
Expand Down Expand Up @@ -38785,8 +38785,8 @@
},
{
"path": "evidence/artifacts/token-adoption-e2e-20260926/README.md",
"sha256": "2849838d915afa5a58bedd9d01a6a91bbd5e416d86a9b91c33d04bba202969d9",
"bytes": 70927
"sha256": "a3bc6729bc53859782337027c8af22f64ebfe276104d54907c580dad36c0b13b",
"bytes": 76920
},
{
"path": "evidence/artifacts/token-adoption-e2e-20260926/RUNBOOK.md",
Expand Down
Loading