Disposition the surface watch's unreviewed names at Claude Code 2.1.300 - #1131

Scope

• What this PR changes: six disposition rows that the first daily run of the newly installed upstream-surface watch listed as unreviewed at Claude Code 2.1.300. These are four new environment variables and two documentation pages whose bodies changed.
• Base commit: 929bc3c3f3b6d1f890279ff14d3f500e2d4db592, the merge base. Corrected after the GPT full1 read; a339425c is an earlier main commit in its history.
• Lane: lane:foundation
• Owned paths touched: catalogs/foundation/upstream-surface-dispositions.json, manifests/evidence.json

SOTA sources

• https://code.claude.com/docs/en/env-vars (env-vars.md read 2026-10-15T01:39:56Z, sha256 54351937…). Rows for CLAUDE_CODE_ARTIFACT_PUBLISH_EDITS (v2.1.300), CLAUDE_CODE_DISABLE_FILE_MENTIONS (v2.1.299), CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS (v2.1.290+, default 200000, floor 100000) and CLAUDE_CODE_YOU_SHOULD_KNOW_DEBUG (v2.1.287).
• The CHANGELOG at anthropics/claude-code 94506ff0d3bd9185f1752da36e968775021fb48f, sections 2.1.299 and 2.1.300.
• https://code.claude.com/docs/en/memory and https://code.claude.com/docs/en/skills, as the watch fetched them at 2026-10-15T01:33:08Z (sha256 1ebf1bdc… and 4d090296…). The memory page still documents @path imports, the AGENTS.md load table and auto memory's first 200 lines or 25KB. The skills page still re-attaches each invoked skill after compaction at its first 5,000 tokens, within a combined 25,000.

Evidence-class table

Local commands run

$ python3 -m unittest tests.test_currency_due tests.test_currency_due_notice tests.test_validate tests.test_evidence_manifest
95 OK; 22 OK; 86 OK; 5 OK
$ UPSTREAM_SURFACE_TEST_CACHE=<empty dir> python3 -m unittest tests.test_upstream_surface_watch
Ran 116 tests, OK (skipped=6)
$ python3 -m unittest tests.test_upstream_surface_watch   # with this host's cache, created by the units installed at 01:32Z
FAILED (failures=4), the four CachedArtifactTests; origin/main fails identically on this host
$ python3 scripts/component_matrix.py --check; python3 scripts/new_host_grand_list.py --check
checked; passed

Decision record

No new decision. The rows follow the dispositions catalog's own rules and the surface-watch guide (docs/upstream-surface-watch.md).

Notes for the reads

• The four host-cache test failures are not from this change. CachedArtifactTests read optional real vendor artifacts from the host's state directory. They skip when it is absent, as in CI and on this host until the daily units were installed tonight. They pin counts from an older vintage of those artifacts: 171 and 173 settings, 1034 Codex config keys. Today's bodies have 172, 174 and 1057. The follow-up is to re-baseline the watch to 2.1.300 and tie those counts to the baseline. That is a separate change for the command center to rule on.
• Not in this PR: a baseline update. The baseline stays 2.1.296 until the follow-up above.

Follow-ups

From GPT full1 at c986f4a (full1-pr1131/GPT-VERDICT-c986f4ac3.md):

• P2, inherited: the optional real-cache tests assert obsolete artifact counts (tests/test_upstream_surface_watch.py:879, :892, :903, :939). Re-baseline the upstream surface watch at Claude Code 2.1.300; hold cached-artifact checks to the baseline #1155's re-baseline closes them, so Re-baseline the upstream surface watch at Claude Code 2.1.300; hold cached-artifact checks to the baseline #1155 lands first and this PR rebases onto it.
• P3: retain the cited source captures and give line locators.
• P3: the host-applicability assertion has no retained readback.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets printed, logged or committed.
• No paid hosting, subscription or billing surface.
• Peer-owned files and worktrees preserved.

🤖 Generated with Claude Code