jericho

Every PIKACI_* environment variable constant across the executor, model, and run modules has been renamed to JERICHOCI_*. This includes:

• Tart VM configuration: JERICHOCI_TART_BASE_VM, JERICHOCI_TART_USE_HOST_XCODE, JERICHOCI_TART_XCODE_APP
• Remote Linux VM settings: JERICHOCI_REMOTE_LINUX_VM_BACKEND, JERICHOCI_REMOTE_LINUX_VM_INCUS_GUEST_ROLE, JERICHOCI_REMOTE_LINUX_VM_INCUS_PROJECT, JERICHOCI_REMOTE_LINUX_VM_INCUS_PROFILE, JERICHOCI_REMOTE_LINUX_VM_INCUS_IMAGE_ALIAS
• Prepared output fulfillment: JERICHOCI_PREPARED_OUTPUT_FULFILL_SSH_BINARY, JERICHOCI_PREPARED_OUTPUT_FULFILL_SSH_HOST, JERICHOCI_PREPARED_OUTPUT_FULFILL_SSH_REMOTE_WORK_DIR, and many more
• Host-local nix shell: JERICHOCI_HOST_LOCAL_NIX_SHELL

All log prefixes have also been changed from [pikaci] to [jerichoci] for consistency in log output. The Tart VM name prefix changes from pikaci- to jerichoci-, and the Incus instance name prefix follows suit. Error messages now reference "Jericho CI" instead of "pikaci Wave 1".

Three new constants govern SSH session resilience:

const SSH_CONNECT_TIMEOUT_SECS: &str = "5";
const SSH_SERVER_ALIVE_INTERVAL_SECS: &str = "15";
const SSH_SERVER_ALIVE_COUNT_MAX: &str = "4";

The run_ssh_command function now passes -o ServerAliveInterval=15 and -o ServerAliveCountMax=4 alongside the existing ConnectTimeout and BatchMode options. This means the SSH client will send a keepalive probe every 15 seconds and tolerate up to 4 missed responses (60 seconds of silence) before disconnecting. A dedicated unit test remote_ssh_commands_set_keepalive_options verifies the exact argument vector.

Two new constants control retry behavior:

const REMOTE_SYNC_MAX_ATTEMPTS: usize = 3;
const REMOTE_SYNC_RETRY_DELAY: Duration = Duration::from_secs(2);

The existing sync_directory_to_remote function is refactored into a retry wrapper that delegates to the new sync_directory_to_remote_once function. On failure, the wrapper logs the attempt number and error, sleeps for the retry delay, and tries again up to 3 times. The log line includes the remote host and the specific label (e.g., snapshot vs. payload) to aid debugging:

[jerichoci] sync {label} attempt {n}/{max} failed for {host}: {err}; retrying after 2s

If all attempts fail, the last error is propagated.

Previously, the match in run_remote_linux_vm_job treated (Ok(_outcome), Err(cleanup_err)) as a hard failure, discarding the successful execution result. The new finalize_remote_linux_vm_execution function extracts this logic and changes the behavior:

• (Ok, Ok) — return the successful outcome (unchanged)
• (Err, Ok) — return the execution error (unchanged)
• (Ok, Err) — NEW: log the cleanup failure to the host log but return the successful outcome
• (Err, Err) — return the execution error with cleanup context (unchanged)

The key insight is that once the guest job has succeeded and produced its artifacts, a cleanup failure (e.g., failing to delete the ephemeral VM) is an infrastructure nuisance, not a job failure. A new test successful_remote_linux_vm_execution_ignores_cleanup_failure validates this behavior.

A new predicate jobs_require_remote_staged_linux_fulfillment checks whether any job in the run uses a remote staged Linux execution path. When it returns true, all eight resolve_run_prepared_output_* functions now receive the jobs slice and bypass recorded metadata to resolve values directly from the live environment:

1. Wrapper program — calls prepared_output_fulfillment_launcher_program() instead of using recorded value
2. Transport mode — forces SshLauncherTransportV1
3. Transport program — reads JERICHOCI_PREPARED_OUTPUT_FULFILL_SSH_BINARY or defaults
4. Transport host — calls prepared_output_ssh_host() directly
5. Remote launcher program — calls prepared_output_remote_launcher_binary()
6. Remote helper program — calls prepared_output_remote_helper_binary()
7. Remote work dir — calls prepared_output_remote_work_dir() directly

This "hard-cut" approach means operators can update infrastructure (e.g., move to a new build host) and have it take effect on the next run without needing to re-record fulfillment metadata.

The IncusGuestRole enum variant PikaciRunner is renamed to JerichoRunner throughout the codebase. This affects:

• The pika-incus-guest-role crate's enum definition and serialization (now serializes as "jericho-runner")
• All test fixtures and assertions that reference the guest role
• Default image aliases that shift from pikaci/dev to jericho/dev
• The Incus run binary path changes from /run/current-system/sw/bin/pikaci-incus-run to /run/current-system/sw/bin/jerichoci-incus-run
• Error messages that mention valid guest role values
• The Nix roles definition in nix/incus/roles.nix

Deserialization remains backward-compatible through the role alias on the image record struct, allowing older records with "pikaci-runner" to still be decoded.

Two new SQL migrations are added:

• 0027: Renames ci_lane_metadata → ci_lane_run_metadata in the ci_lane_states table
• 0028: Renames structured_ci_target → ci_target in the ci_lane_states table

The corresponding Rust code in ci_store.rs updates all SQL queries to use the new column names. The CiLaneState struct fields and the CiManifest / CiState modules are updated accordingly to use ci_lane_run_metadata and ci_target consistently. The ci_manifest.rs module also renames CiTarget references and the ci.rs module adjusts its lane state construction.

A new jerichoci_store module is added under pika-git/src/ and registered in the crate's module tree via storage.rs. The branch store is updated to reference CI state through the new Jericho-named structures. This module provides the storage layer for Jericho CI run metadata, working alongside the existing ci_store module which handles the renamed columns from the database migrations.

The Nix and scripting infrastructure receives matching updates:

• nix/incus/pikaci-image.nix — Image alias references shift to jericho/dev pattern; the guest image Nix expression is updated
• nix/incus/roles.nix — Guest role definitions updated from pikaci-runner to jericho-runner
• nix/ci/linux-rust.nix — CI derivation references updated
• scripts/pikaci-staged-linux-remote.sh — Remote staging script updated with new paths (/var/tmp/jerichoci-prepared-output) and environment variable names
• scripts/lib/pikaci-tools.sh — Shared shell library updated for new variable names
• scripts/pikaci-incus-image.sh — Incus image build/import script references updated
• flake.nix — Nix flake outputs adjusted for renamed derivations
• infra/justfile and just/checks.just — Just recipes updated to reference new binary and path names

The .gitignore also adds .jerichoci/ to ignore the new local working directory.

The web layer is updated across views, API handlers, and HTML templates:

• web/api.rs — API response structures use the renamed ci_lane_run_metadata and ci_target fields
• web/views.rs — View models updated to match the renamed CI state fields
• branch_ci_live.html and nightly_live.html — Templates reference updated field names for rendering CI lane states
• Test modules (api.rs, live.rs, render.rs, unit.rs) — All test fixtures and assertions updated to use the new naming. This includes JSON snapshot assertions and rendered HTML checks.

The web tests are particularly important as they serve as integration-level verification that the renamed columns flow correctly from the database through the store layer to the API and rendered views.

The forge model crate and pikaci orchestrator are updated:

• pika-forge-model — The CI target struct and its serialization are updated to use the renamed ci_target field name
• pikaci/src/targets.rs — Target resolution logic references updated field and type names
• pikaci/src/ci_catalog.rs — CI catalog definitions updated for Jericho naming
• pikaci/src/forge_lanes.rs — Forge lane construction uses renamed types
• pikaci/src/main.rs and pikaci/src/lib.rs — Entry points and library root updated
• pika-git/src/forge.rs — Forge integration store queries use the renamed columns

These changes ensure that when the forge (e.g., GitHub/GitLab integration) triggers CI lanes, the data flows through with consistent Jericho naming from target resolution through to state persistence.

The cloud and server layers receive targeted updates:

• pika-cloud/src/incus.rs — Incus VM provisioning references the renamed JerichoRunner guest role and updated image aliases
• pika-cloud/src/spec.rs — Cloud spec definitions use the new role names
• pika-server/src/agent_api.rs — The agent API that CI runners call back to is updated to handle the renamed CI metadata fields

These changes ensure that the full lifecycle — from cloud VM provisioning through agent registration and result reporting — uses consistent Jericho naming.

Supporting files are updated for completeness:

• ph/src/commands.rs and ph/src/tests.rs — The ph CLI tool's CI-related commands reference updated types and field names
• jerichoci/src/snapshot.rs — The snapshot metadata marker file is renamed from pikaci-snapshot.json to jerichoci-snapshot.json, and all path references (temporary directories, remote staging directories) use the jerichoci- prefix
• pika-git/README.md — Documentation updated to reference Jericho CI
• Todo documents (jericho.md, pika-cloud.md, pika-git.md, professional-infra.md) — Planning documents updated
• .gitignore — Adds .jerichoci/ to the ignore list alongside the existing .pikaci/ entry
• crates/pikahut/tests/guardrails.rs — Guardrail tests updated for the new naming