DO NOT MERGE — merge train: 3016,3017,3018,3025,3008 - #3027
Closed
trinity-ability wants to merge 16 commits into
Closed
trinity-ability wants to merge 16 commits into
trinity-ability wants to merge 16 commits into
Conversation
…ivery failures reach REST and MCP callers (#2991) `start_agent_internal` computed the skill-delivery result, and `POST /api/agents/{name}/start` rebuilt its response from a whitelist that left it out, so a caller could not tell a clean start from one whose assigned skill failed or hit a name conflict. - The endpoint returns `skills_injection` + `skills_result` beside the credentials pair. `skills_result` is a projection (`public_skills_result`): names, statuses and codes only — a per-skill error that carries an exception string, URL or path is reduced to a code (ent#334). A no-op start answers `skipped` + reason; a missing result reads `unknown`, never absent. - A conflict keeps `success` (#2914), so `inject_assigned_skills` now names it in `conflicts` — distinguishable from a clean start in one response. - MCP `start_agent` keeps its first line and, when delivery was not clean, adds the status and one line per skill that did not land. - Same gap closed on `POST /api/system-agent/restart`; the fleet restart entry gains `skills_conflicts`. The projection is shape-defensive: it runs inside the start's try, so a raise would turn a start that happened into a 500. Mutations: whitelist reverted (6 red), raw result passed through (5 red), lifecycle drops conflicts (1 red), MCP returns only the message (2 red). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…efuses on a shared source-mode branch (#3011) The heartbeat was add -A -> commit -> push origin HEAD with no fetch, so the first foreign push to the agent's branch failed every later cycle non-fast-forward, forever, with the agent's commits piling up locally. - fetch origin <branch>; if behind, rebase --autostash onto it and push with --force-with-lease pinned to the fetched sha (a racing push is rejected, never overwritten); a plain push otherwise; a branch not yet on the remote is created by the push - rebase conflict (or timeout): rebase --abort, repo left as it was, record 'diverged: rebase conflict on <branch>' for the existing sync_failing path; never resolve, overwrite or reset - source-mode agent on the default branch (origin/HEAD, else main/master) refuses before committing: 'refused: source-mode on <branch>'; fork-to-own agents (GIT_UPSTREAM_REPO or an upstream remote) exempt - sync-state.json gains behind_after_fetch and last_successful_push_at Fixes #3011 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
PUT /git/auto-sync wrote the DB row and nothing else: the container gated on GIT_SYNC_AUTO read once at startup, and every recreate re-derived it as DB flag OR baked env. OFF never took effect on an agent that baked the env at creation; ON waited for the next recreate. - agent loop: starts whenever it can ask the platform and reads the owner's flag every cycle via GET /git/auto-sync with its own MCP key; OFF skips the cycle, ON runs it, no recreate. 404 'Git not configured' -> off; platform unreachable / 5xx / auth refusal -> the env fallback. /api/git/status reports the value the loop runs with. - one writer: _apply_git_env_from_db derives GIT_SYNC_AUTO from the DB flag alone (OR + log removed); creation writes the flag from the same _git_auto_sync_baked predicate that bakes the env, ghosts included. - one-shot backfill (SQLite + Alembic 0075): live non-source-mode ghosts, the env-true/DB-0 slice the DB can identify. - Settings -> Git sync panel with both toggles (auto-sync, pause schedules while sync is failing). Fixes #3010 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…omments (#3010) Found on a live dev agent: the first cycle waits a full interval, so /api/git/status reported the env fallback (off) for 15 minutes while the owner's flag said on. Resolve once before the first sleep. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A Contents: read-only fine-grained PAT passed every check (ls-remote talks to upload-pack; REST permissions.push is the user's role, not the token's grant) and then failed every auto-sync for the agent's life. - git_service.probe_push_access: git push --dry-run of a throwaway ref from an empty scratch repo, PAT via git_auth_env (never argv); asks receive-pack, creates nothing. ok / denied / transient. - creation: agents that will auto-push (_git_auto_sync_baked) are probed; denied -> 400 naming GitHub's reason and the fix, before any branch is reserved or container created; transient -> logged, non-blocking. - sync health: a sync_failing item whose error is a refused push is titled 'Git token can't push', says it won't recover on its own, and carries cause=push_denied + remediation. Fixes #2107 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…2107) CodeQL py/clear-text-logging-sensitive-data: the line comes from a child that held the token. It is still returned (scrubbed) to the caller. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…2105) `_dual_ahead_behind_payload` measured the working tuple against `origin/<current_branch>` only for `trinity/*` branches. Every other branch got the `origin/main` counts under the working label. The fleet audit reads `ahead_working` as unpushed commits, so a fleet renamed to `asus/*` reported ~9,000 unpushed commits while the true gap was 0. - The working tuple uses `origin/<current_branch>` whatever the branch is named. - With no upstream (never pushed, detached HEAD), `ahead_working` counts the commits no remote holds, and `behind_working` is null. - A count that can't be computed is null, never 0. This covers the main tuple on a `master` repo with no `main`, which read 0 ahead for 7 days while one commit sat unpushed. The backend already stores null as 0, and the UI checks `> 0`. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…mechanical, per the merge-train note on the PR `ahead` aliases the main tuple, which this PR makes None on a repo with no `main`; `None == 0` is false, so a clean `master` repo flipped from "Synced" to "Changes pending". Restores dev's behaviour on that line. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This was referenced Sep 25, 2026
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
3 participants
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.
Integration surface for #3016, #3017, #3018, #3025, #3008. Never merged; members merge individually once green.
🤖 Generated with Claude Code