Token-profile install receipts and a MCPorter use receipt on the workstation (MCPorter host_verified) - #277

Scope

• What this PR changes: six host receipts on the workstation nativestack-5975wx-20260925, closing the token-efficiency profile's receipt gaps that remained after Record token-stack host receipts on the workstation: RTK hook path, repomix, toon, Context Mode, jCodeMunch, Serena, ccusage #263:

  • install receipts for context-mode, Serena, jCodeMunch, SocratiCode and MarkItDown;
  • a native_proven MCPorter use receipt against a real MCP server.

  All six are at the current winner or stack pins. Those pins come from the sealed 2026-09-22 packets. A pin move in stack.json or the pins files does not retire these receipts until a new verdict wave, as cleanup-wsl-audit confirmed on vLLM 0.30.0 host receipts on the workstation: install agree, use pending the winner pin #267.

• Base commit: 5b6014eb (origin/main). The receipts were recorded at dcd6f0cf, which is each receipt's catalog_revision and a published ancestor of the base. The branch was rebuilt on 5b6014eb under the hot-file protocol: main's manifests/evidence.json was taken, the six files were re-registered, and the matrix and grand list were regenerated.

• Lane: lane:foundation

• Owned paths touched: evidence/hosts/nativestack-5975wx-20260925/ (6 new receipts). Regenerated: catalogs/landscape/component-evidence-matrix.json, docs/component-evidence-matrix.md, catalogs/landscape/new-host-grand-list.json, docs/new-host-grand-list.md. Hot file: manifests/evidence.json, updated in the last commit.

Receipts (generated from the receipt JSON)

Recorder and reviewer identities differ. The recorder, 106d189e…, is the coordinator session. The reviewers (ef0949f4… for the first round, then 6e198bff… and 2719d754… for the MCPorter generations) are separate headless claude -p session on claude-sonnet-5. It could not edit files, and it appended its verdicts through scripts/host_receipts.py review --kind independent_session. Each review's --ref names the commands it re-ran and the points it checked.

Effect on the derived views (generated)

• MCPorter moves from accepted to host_verified in the grand list. The WSL2 pairs that are host_verified on a host receipt go from 27 to 28. The WSL2 needs_host count stays at 18.
• context-mode and jCodeMunch are not layer winners. Their install receipts bind only to the manifests/stack.json pin, and they complete the token-efficiency adoption profile (Token-efficiency coverage check: adoption profile plus value-free --client-wiring for every host #264).

Repair round (after the independent diff review)

The read-only diff review found that the first MCPorter use receipt let QMD's default reranking load a model over the network into the temporary HOME, and the limitations didn't say so. Receipts are immutable, so the fix is a superseding generation, recorded from a detached worktree at published origin/main 5b6014eb:

• -2: disables rerank and runs the server calls under bwrap --unshare-net. Its own independent review held it at needs_changes on point 3: the "QMD 2.8.3" and "no host mcporter configuration" limitations had no retained command, and mcporter ran with the real HOME.
• -3: adds qmd --version and runs mcporter itself under the temporary HOME. An independent re-check reproduced all four output hashes and ran a config-canary control, and it agreed on all four points.

The first and -2 receipts stay in the tree as superseded history. Only -3 counts.

How each receipt avoids known failure modes

• MCP Inspector never runs (it would open a browser on this host). MCPorter (-3) bridges a fresh QMD stdio server with networking unshared and runs under a temporary HOME, never a production server or the host's mcporter config.
• Positive controls:
  • MCPorter: a nonexistent-tool call exits non-zero with a "not found" error, and a miss query returns 0 results.
  • jCodeMunch: a deliberate one-file append is detected as a RECORD mismatch.
• Integrity limits stated in each receipt:
  • SocratiCode has no --version, so its version comes from npm metadata.
  • MarkItDown's adoption sha256 is for the sdist, which is not comparable to the installed RECORD.
  • Serena and jCodeMunch have no Linux pins-file entry.

Local commands run (on this branch)

scripts/component_matrix.py --write; scripts/new_host_grand_list.py --write
the offline steps of validate.yml's validate job, in order (zizmor and actionlint run in CI only; verdict_flip_candidates is report-only), plus evidence_manifest.py --check (a files[] order check that CI runs indirectly through validate.py): workflows SHA256SUMS + node tests, validate.py, host_receipts.py validate,
  release_due.py --strict-if-repinned origin/main, validate_catalogs.py, validate_foundation.py, landscape.py,
  audit_reports.py --check, validate_convergence.py --all-recorded, build_verdicts.py --check,
  component_matrix.py --check, new_host_grand_list.py --check, gap_crosswalk.py build --check,
  gap_wave_ledger.py --check, build_ecosystem.py --check, evidence_manifest.py --check   -> all passed (0 failures)
verdict_review_gate.py (from a worktree at origin/main) --root . --base origin/main -> passed (0 changed rows, 0 violations), on both bases
python -m unittest (uv Python 3.13 with jsonschema/PyYAML/numpy; PROMOTION_GATE_PYTHON set) on base dcd6f0cf -> Ran 5367 tests, OK (skipped=612)
after the rebase onto 5b6014eb and after the repair commit: the validate list above again (0 failures) + tests.test_host_receipts/component_matrix/new_host_grand_list/landscape/adoption_status/evidence_manifest/validate* -> Ran 441, OK

Residuals

• Codex cross-family review skipped: the Codex usage limit resets 2026-09-30 23:50 local. The Opus weekly limit resets 2026-09-30 21:00 ET, so the reviewer ran on Sonnet.
• Receipt gaps still open:
  • A Codex use receipt waits for the Codex reset.
  • Claude Code use waits: the winner pin is 2.1.278, the floor 2.1.281 and the host runs 2.1.282, so this belongs in the wave.
• Coverage from a fresh worktree: Codex MCP servers are wired only in the main checkout's excluded .codex/config.toml, so the token-efficiency coverage check is not complete from a fresh worktree. A follow-up PR, after Token-efficiency coverage check: adoption profile plus value-free --client-wiring for every host #264 and Token report: take the Codex hook engine from the stable hooks flag; drop the removed plugin_hooks step #265, fixes that and adds a pinned-version check.

🤖 Generated with Claude Code