Repository navigation
Description
Part of #14559 (per-backend split of the backend-specific routes). Epic: #14550. Target: v6 (open the PR with --base v6).
Scope (C1, C3, C14): copilot runs one long-lived Bash session per shell id, so the neutral BASH_ENV route never sees its commands. Add a producer (route e.g. copilot_events) that reads the Copilot session log ~/.copilot/session-state/<id>/events.jsonl (per-agent COPILOT_HOME, provisioned by setupCopilotHome in src/crux/agent/interactive_home.go) and emits one shared command record per shell tool.execution_complete event: command from the matching tool.execution_start / toolRequests arguments, result from data.success (Failed = !success; set ExitCode only if the event carries a numeric exit code — check). Reuse the event struct and parsing in renderCopilotTranscript (src/crux/agent/kick_transcript_copilot.go) and the per-session discovery in newestCopilotTranscriptActivity (src/crux/agent/busy_visibility.go); emit exactly one record per tool call across re-reads (track offset / toolCallId). Shell-activity signal for C12: shell tool.execution_start events.
Only write into copilot's settings files where Hive already writes them (the advisor hooks dir under the per-agent .copilot); with repeat_failure.enabled off Hive adds nothing and removes what it added at the next launch (C14 / C-AC12).
Shared pieces (v6 @ 6cec2b9): record type RepeatFailureRecord (Route, ExitCode *int, Failed) and writer AppendRepeatFailureRecord / RepeatFailureRecordPath(home) in src/crux/agent/repeat_failure_record.go; switch via Manager.repeatFailureEnabled() and route-constant pattern RepeatFailureRouteBashEnv = "bash_env" in src/crux/agent/repeat_failure_bash_env.go. Coverage table repeatFailureCoverageTable in src/crux/agent/repeat_failure_coverage.go; tests TestRepeatFailureCoverageRowsAreHonest / TestRepeatFailureCoverageDocsMatchCode; docs table "Repeated-failure check" in src/docs/backend-support-tiers.md (~line 550, rendered from RepeatFailureCoverageMarkdownRows()).
Harness (C4): a committed, re-runnable harness (stub model provider that makes the real copilot CLI run false or ls /nonexistent once, or a recorded real-run events.jsonl fixture) with the copilot --version output captured, proving one failing record. Update the copilot row in repeatFailureCoverageTable (Covered, Route, Evidence incl. --version, EvidencePath, ShellActivity) and the matching docs row.
If it cannot be covered (C5): leave the row "not covered", put what was tried in its Reason, and keep the docs row in sync.
Acceptance: C-AC1 for copilot when covered, else C-AC9; the C-AC11 coverage test stays green.
Decision taken by hive: split #14559 per backend so each lands as one reviewable PR; no ordering dependency between backends.
— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88
Activity
hivecommons-hive commentedon Oct 28, 2026
Folded finding. This was filed as a separate issue, but 2 open agent-filed issues (plus 1 closed as completed in the last 14 days) already cite
src/docs/backend-support-tiers.md, so it was folded into the oldest of them instead of opening another. Treat this issue as the tracker for that component: prefer one structural fix that covers every folded variant over a PR per variant.[scanner] repeat-failure: openhands route emits the shared command record + re-runnable harness
Part of #14559 (per-backend split of the backend-specific routes). Epic: #14550. Target: v6 (open the PR with --base v6).
Scope (C1, C3, C14): openhands keeps one interactive
/bin/bashopen in tmux, so the neutralBASH_ENVroute never sees its commands. Planned routes: aPostToolUsehook (~/.openhands/hooks.json; check the hook input fields) or a reader of its saved events (exit_code), emitting one shared command record per shell command (route e.g.openhands_hook/openhands_events).Hook wiring location TBD — not found on v6 at 2026-10-27: openhands has no Hive hook wiring or launch case on v6 (it falls to the
defaultbranch ofbackendLaunchCmdWithBobSessioninsrc/crux/agent/manager_launch.go; onlyconfig.CLIBackendsinsrc/crux/config/agent_runtime.goand the coverage row name it), and Hive does not write~/.openhands/hooks.jsontoday. So prefer a Hive-side reader of the saved events, or a launch-env / Hive-owned-file route; writinghooks.jsonis only allowed if Hive already writes that file for openhands. Withrepeat_failure.enabledoff Hive adds nothing and removes what it added at the next launch (C14 / C-AC12).Shared pieces (v6 @ 6cec2b9): record type
RepeatFailureRecord(Route,ExitCode *int,Failed) and writerAppendRepeatFailureRecord/RepeatFailureRecordPath(home)insrc/crux/agent/repeat_failure_record.go; switch viaManager.repeatFailureEnabled()insrc/crux/agent/repeat_failure_bash_env.go. Coverage tablerepeatFailureCoverageTableinsrc/crux/agent/repeat_failure_coverage.go; testsTestRepeatFailureCoverageRowsAreHonest/TestRepeatFailureCoverageDocsMatchCode; docs table "Repeated-failure check" insrc/docs/backend-support-tiers.md(~line 550).Harness (C4): a committed, re-runnable harness (stub model provider driving the real openhands CLI, or a recorded real-run fixture) that records the
openhands --versionthe harness installs (openhands is not in the image), showing one failing command recorded. Update the openhands row inrepeatFailureCoverageTable(Covered, Route, Evidence incl.--version, EvidencePath, ShellActivity) and the matching docs row.If it cannot be covered (C5): leave the row "not covered", put what was tried in its
Reason, and keep the docs row in sync.Acceptance: C-AC1 for openhands when covered, else C-AC9; the C-AC11 coverage test stays green.
Decision taken by hive: split #14559 per backend so each lands as one reviewable PR; no ordering dependency between backends.
— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88
hivecommons-hive commentedon Oct 28, 2026
The folded openhands finding above was filed separately as #14748. This issue covers copilot only.
— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88
hivecommons-hive commentedon Oct 28, 2026
🔒 Claimed by agent
scanner(agent on hive hosted-available-oke-11-placeholder-r05x). The claim lasts 2h0m0s without activity and renews while the work continues. Reply/unclaimto release, or/claimto take it over if you outrank the holder.hivecommons-hive commentedon Oct 28, 2026
🔓 Claim by agent
scannerreleased (holding agent relaunched).2 remaining items
hivecommons-hive commentedon Oct 28, 2026
🔓 Claim by @thatguywob released (lease revoked).
hivecommons-hive commentedon Oct 28, 2026
🔒 Claimed by agent
scanner(agent on hive hosted-available-oke-11-placeholder-r05x). The claim lasts 2h0m0s without activity and renews while the work continues. Reply/unclaimto release, or/claimto take it over if you outrank the holder.github-actions commentedon Oct 28, 2026
Closed by #14801 (merged to
v6).