memory: record the PR #2 promotion and correct the false prior-claim entry - #3
Open
compusophy wants to merge 10 commits into
Open
memory: record the PR #2 promotion and correct the false prior-claim entry#3compusophy wants to merge 10 commits into
compusophy wants to merge 10 commits into
Conversation
0582489 failed BOTH gates. this fixes the diagnosed cause and ships the diagnostic capability whose absence made the failure undiagnosable: 1. ci workflow: cargo check --lib --bins on wasm32 (NOT --all-targets). --all-targets pulls integration tests onto wasm32-unknown-unknown, where the `test` crate is not shipped — instant E0463 on every run regardless of code health. that is what killed the workflow's first run in 43s at step 5. pinned by tests/ci_gate.rs:: the_wasm_check_never_builds_test_crates. 2. gate diagnostics ride $GITHUB_STEP_SUMMARY (ci/run_tests.sh + a summary-writing step in the workflow). job LOGS need admin over the api; the check-run SUMMARY is public payload. next red build is readable by the agent itself: fetch check-runs → read output.summary → see actual rustc lines instead of "exit code 101". 3. worker self-config from opfs (the user's design question, answered in code): ui save_config mirrors Config to vanish-config/config.json; boot_worker reads it and runs the full Configure path (credential verification + D10 auto-reconcile) itself. localStorage is ui-thread-only, which is why the worker booted credential-blind and could not read its own build logs even though the vercel token was saved in the panel. memory records both causes and the still-open question (whether the vercel/native failure was ONLY the wasm lib or also the never-compiled eval suites — the summary will say if not).
summary api returned null, annotations carry only exit codes, job logs need admin: none of the three gave the agent its own compiler output. now ANY failure pushes ci-diagnostics.log (wasm check tail + every gate suite tail) to the diagnostics branch, readable unauthenticated at raw.githubusercontent.com/<repo>/diagnostics/ci-diagnostics.log. also: wasm check output captured via tee+PIPESTATUS so a failing cargo check still leaves its full log behind; permissions block grants contents:write for the publish push. no source change — this commit exists to make the next red build readable by the thing repairing it.
root cause of the three red builds, straight from the diagnostics log: the committed control.rs was a stale partial copy ending at history_is_well_formed — the whole automatic-continuation section (LoopContinuation, decide_after_run_end, resume_marker_is_fresh, RESUME_MARKER_MAX_AGE_MS, RESTART_WINDOW_MS, MAX_RESTARTS_PER_WINDOW, RestartBudget) existed in the working tree but never reached github, while worker.rs DID ship complete and references every one of those items: 10 E0425/E0433 in the wasm lib on every platform. the read-through cache serving the complete file while the committed bytes were partial is incident class #4 of the stale-tree family; the diagnostics branch is what made it visible — one push and the exact rustc errors were readable unauthenticated. that pipeline stays.
both failures were in never-compiled code from the token-blocked session: 1. test_count_and_commit_checkers_read_the_snapshot asserted CommitExists passes while has_commit was still false — the field was simply never set true first. now set explicitly. 2. file_excludes_fails_when_token_survives_anywhere pinned "a deleted file passes an excludes checker", contradicting both the impl (which deliberately requires existence) and the suite invariant that an empty tree scores zero. a removal task graded as success after rm-ing the whole file is gameable benchmark design; the pin now asserts the strict semantics and says why.
the diagnostics log caught TWO more stale-commit casualties and one of my own mistakes: 1. committed build.sh was the OLD hand-rolled version (for suite in protocol_contract ... agent_evals — omitting bench_grading and ci_gate, the exact rot the shared gate exists to prevent). restored to delegation: bash ./ci/run_tests.sh. both guard tests now pass by construction. 2. my own new guard test tripped on itself: it greps the workflow for a flag literal that also appeared in its own doc comment and in the workflow's comment text. grep guards must not name their needle in prose; the literal is now assembled from fragments so neither this file nor the workflow contains it. bench_grading 9/9 green in this run; remaining failures were ci_gate only. wasm lib check green for the second consecutive run.
every test suite is now green (agent_evals 15, bench_grading 9, branch_policy 7, ci_gate 5, event_loop_liveness 5, loop_nervous_system 9, platform_logic 25, protocol_contract 11, streaming 11 — all passing in the diagnostics log for 27a3506). the only remaining failure was clippy::single_match on the outer read() match whose Err arm was empty; restructured to if-let per the lint.
…le-partial-copy incident class
…k is green at bba0104/19c92af
…le constitution, wired into boot order the owner set the goal explicitly: agi, rsi, asi. until now intent lived nowhere — architecture carried the how, directives carried constraints, memory carried tactics, nothing carried purpose. a self-editing system reconstructing its purpose from whatever is broken at boot drifts; the charter is the fixed point. contents: the mission (recursive self-improvement, vanish as self-sovereign vehicle), eight articles derived from existing D-law, honest current-state (not agi, weak rsi with human-steered objectives), measures of progress that can fail, amendment clause (owner enacts, agent proposes only). wired into boot order: system prompt reads charter first; taskboard header carries read order (charter → taskboard → status). landed on the repaired green base so promotion carries only this. (the same content briefly existed on an orphaned agent/charter branch rooted at the pre-fix base; that pointer is superseded by this commit.)
…ol, and CHARTER are on main TASKBOARD: token-scope item closed with the complete fine-grained scope list (Contents rw, Workflows rw, Pull requests rw — Workflows alone was never enough); the false "promoted through a green pr" claim about 0582489 corrected (that landing went red; the superseding branch landed as PR #2). operational notes recorded: open_pr refuses from main — checkout the agent/ branch first; a page reload resets the session branch to main. evidence: compare api showed ahead_by 9 / behind_by 0 before opening; both gates green at 4ee87bc; merged as squash e48e4ee on main. status.md carries the full run record including the D10 guard firing correctly on my own commit and the orphaned-commit recovery. changeset is memory-only (TASKBOARD.md + status.md): no source touched, no build impact possible.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
compusophy
added a commit
that referenced
this pull request
Aug 25, 2026
… after conflict) TASKBOARD: token-scope item closed with the complete fine-grained scope list; the false "promoted through a green pr" claim about 0582489 corrected (it went red and never merged; the superseding branch landed as PR #2, squash e48e4ee). status.md: full run record — the 403 → scope fix → PR #2 green merge chain, the strictly-ahead compare check before opening, open_pr's refusal from main, D10 firing correctly on my own second commit, and the created_at staleness wrinkle in git_create_branch. first attempt at this changeset (PR #3 from a pre-merge base) conflicted by construction — memory files were rewritten by the very merge being recorded. rebased onto e48e4ee instead of force-resolving. memory-only changeset: no source touched, no build impact possible.
compusophy
added a commit
that referenced
this pull request
Aug 25, 2026
… after conflict) (#4) TASKBOARD: token-scope item closed with the complete fine-grained scope list; the false "promoted through a green pr" claim about 0582489 corrected (it went red and never merged; the superseding branch landed as PR #2, squash e48e4ee). status.md: full run record — the 403 → scope fix → PR #2 green merge chain, the strictly-ahead compare check before opening, open_pr's refusal from main, D10 firing correctly on my own second commit, and the created_at staleness wrinkle in git_create_branch. first attempt at this changeset (PR #3 from a pre-merge base) conflicted by construction — memory files were rewritten by the very merge being recorded. rebased onto e48e4ee instead of force-resolving. memory-only changeset: no source touched, no build impact possible.
compusophy
added a commit
that referenced
this pull request
Aug 25, 2026
both report mergeable=None after today's merges. their unique content (observability roadmap: runtime error capture, ui snapshot-as-text, boot reconcile retry) is now on the board directly so closing the PRs cannot lose it — a memory claim that exists only in an unmerged branch is one conflict away from vanishing (d846fcd class).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
memory-only changeset (TASKBOARD.md + status.md); no source touched.
records on the permanent record:
restored loop-survival control, worker self-config, CHARTER now live
on main.
Workflows rw, Pull requests rw, Metadata read; Checks/Actions read) —
Workflows alone was never sufficient.
promoted: it went red on both gates and was never merged; found by
observing main still lacked ci/run_tests.sh entirely.
page reload resets session branch to main; D10 refusal fired once
during this very run and the reconcile-retry path worked as designed.
observable behavior: none at runtime — this is the memory layer the
next run reads first.