Skip to content

Repin EndoForge to main 68eaa75 (macOS durable sync) and make run_status honest - #88

Merged
Akilleez-QA merged 2 commits into
mainfrom
fix/repin-endoforge-macos
Nov 17, 2026
Merged

Akilleez-QA merged 2 commits into
mainfrom
fix/repin-endoforge-macos

Conversation

@Akilleez-QA

Copy link
Copy Markdown
Owner

What

  1. Repin vendored EndoForge to origin/main 68eaa75a0ce5f65b5517077cb736d8d7ffebb755 (was ea7b8d6), using tools/pin_endoforge.py --source <clone> --commit 68eaa75…. SOURCE-MANIFEST.json and contracts/SOURCE-PINS.json record this commit with on_main: true. The slice grows from 27 to 29 files: it adds vendor/endoforge/tools/durable.py (portable durable sync: F_FULLFSYNC on macOS, else fdatasync, else fsync; endoforge #163), which the ledger writer now imports, and vendor/endoforge/core/runtime/isolate-wasm-tooling.mjs, which IsolatePool and the WASM limit now import (endoforge #153, #155).
    • Why: the old ledger writer called os.fdatasync, which macOS Python does not have. On macOS every forge goal failed with isolate ledger bridge failed: exit 70.
    • Checked against icd-endoforge.md §8: no interface Cstack calls has changed. Ledger and FrozenEvaluators are byte-identical. IsolatePool only adds two optional settings, toolingWasmModules and syncHostCalls, and Cstack sets neither. ENDOFORGE_LEDGER_JOURNAL and ENDOFORGE_DUMP_LIST are read the same way as before.
  2. Honest top-line status in cli.py. run_status is completed only when every outcome executed and responded (completed or tool_error with certainty responded). Otherwise it is one of these, in this precedence: stopped, outcome_unknown, failed, not_executed. A new top-level not_completed list gives decision_id, attempt_id, execution_status, execution_certainty, assessment and limitations for each outcome that did not run. The rule is a new requirement, CS-REQ-088, enforced by tests/test_process_and_cli.py::RunStatusTests, tests/test_forge_identity.py::NotExecuted and tests/test_workcell_run.py::OneGoal.
  3. Docs updated: pin ICD, EndoForge ICD (§1, §8), configuration management, the vendoring explanation, the repin how-to, the command reference (run_status, not_completed), test plan TP-4, requirements, the verification matrix, the SDD-C9 row and CHANGELOG. EXPORT-MANIFEST.json is regenerated.

Checks (run on macOS, Darwin, Node 26.13.1; no model calls; ANTHROPIC_* and CLAUDE_CODE_OAUTH_TOKEN unset; stub transport only)

Check Result
python -B -m unittest discover -s tests, Python 3.13 venv with ci/requirements.txt, venv on PATH as in CI before (origin/main 0a9703b): 734 tests, 71 failures + 3 errors; after: 735 tests, 5 failures, 0 errors
Same suite, Python 3.15.2 venv with ci/requirements.txt (the cold-start probe's setup) before: 70 failures + 4 errors (matches the probe); after: 5 failures, 0 errors
CSTACK_ENDOFORGE_CLONE=<fetched clone> python -B -m unittest tests.test_endoforge_pin tests.test_pin_endoforge_tool 9 tests OK, none skipped
python -B tools/check_trace.py 0 failures (78 requirements, 76/76 test files traced)
python -B tools/check_links.py 0 broken
python -B tools/gen_data_dictionary.py --check up to date
Forge example from first-run.md step 5 (stub transport), before completed workcell_not_executed: the old status claimed success, and the limitations held isolate ledger bridge failed: exit 70
Same example, after (Python 3.15 and 3.13) completed candidate_passed assemblyscript 0.28.20 {'foundation_refrigeration': 'js-port'} unavailable, not_completed: []
Failure path, after (forge.sqlite made a directory) not_executed, with not_completed = unavailable / not_dispatched / workcell_not_executed / "Reason: unable to open database file"
Diff scan for personal data, secrets and new code comments clean (gitleaks and trufflehog are not installed locally; CI runs them)

The before and after suites ran in pairs at the same time, so both had the same load.

Not done

  • The 5 remaining macOS failures are all in tests/test_workcell_kill.py::Kill. They were already failing on main. The test helpers use GNU ps --ppid (in tests/forge_fixtures.py::descendants) and /proc/<pid>/stat (in alive), and macOS has neither, so wait_for times out. The product code is not at fault. A portable helper (pgrep -P, ps -o stat=) would fix it; that is a follow-up.
  • The tutorial transcript in first-run.md was recorded on Linux and was not re-recorded.
  • CS-REQ-088 was chosen after checking open PRs and sibling branches: 080–087, 090 and 091 are taken.

🤖 Generated with Claude Code

Cstack and others added 2 commits November 16, 2026 23:56
…atus

The vendored EndoForge ledger writer called os.fdatasync, which macOS
Python lacks, so every forge goal on macOS failed with "isolate ledger
bridge failed: exit 70". Repin with tools/pin_endoforge.py to EndoForge
origin/main 68eaa75a0ce5f65b5517077cb736d8d7ffebb755, which syncs through
tools/durable.py (F_FULLFSYNC on macOS, else fdatasync, else fsync;
endoforge #163). The slice gains tools/durable.py and
core/runtime/isolate-wasm-tooling.mjs, now imported by the ledger writer
and IsolatePool (27 -> 29 files). No interface in icd-endoforge.md
section 8 changed: Ledger and FrozenEvaluators are byte-identical, and
IsolatePool only gained optional options Cstack does not pass.

cli.py reported run_status "completed" even when the workcell never ran.
It now reports completed only when every outcome executed and responded;
otherwise stopped, outcome_unknown, failed or not_executed, and a new
top-level not_completed list carries each such outcome's status,
assessment and limitations. Enforced by RunStatusTests and the
NotExecuted forge tests under new requirement CS-REQ-088.

Full suite, before -> after: Python 3.13 CI-like venv 71F+3E -> 5F;
Python 3.15 on macOS 70F+4E -> 5F. The remaining five are
tests/test_workcell_kill.py, whose helpers use GNU ps --ppid and /proc.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brings in #84 to #87. Conflicts resolved by keeping both sides:
CHANGELOG.md entries; requirements and verification-matrix revision
rows renumbered so main's 2.2 (cstack init) stays and this PR's
CS-REQ-088 row becomes 2.3, with each document's Version set to 2.3.
EXPORT-MANIFEST.json regenerated with tools/build_export_manifest.py;
data dictionary --check passes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Akilleez-QA
Akilleez-QA marked this pull request as ready for review November 17, 2026 09:58
@Akilleez-QA
Akilleez-QA merged commit b0761b1 into main Nov 17, 2026
2 checks passed
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant