gitea-admin/CASAN
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: prepare CASAN paid PoC release package

Browse Source
This commit is contained in:
thanhnv
2026-07-18 00:07:04 +07:00
parent c818eaf8b0
commit 881ee01691
55 changed files with 5992 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/casan-harness/scripts/bash/ci-harness-gate.sh
+12
View File
@@ -7,6 +7,7 @@ set -uo pipefail
#
# Env:
# CASAN_CI_RUN_FRONTEND=0|1 default 1
# CASAN_CI_RUN_CONTROL_PANEL=0|1 default 1
# CASAN_CI_RUN_INFRA_LAB=0|1 default 0 (Docker Compose lab is optional in CI)
# CASAN_CI_STEP_TIMEOUT_SEC default 600
# CASAN_CI_SUITE_FILTER optional regex; run matching suite names only
@@ -188,6 +189,17 @@ else
skip "frontend-vitest (CASAN_CI_RUN_FRONTEND=0)"
fi
if [[ "${CASAN_CI_RUN_CONTROL_PANEL:-1}" == "1" ]]; then
if command -v npm >/dev/null 2>&1; then
run "control-panel-tests" npm run console:test
run "control-panel-build" npm run console:build
else
skip "control-panel-tests/build (npm unavailable)"
fi
else
skip "control-panel-tests/build (CASAN_CI_RUN_CONTROL_PANEL=0)"
fi
if [[ "${CASAN_CI_RUN_INFRA_LAB:-0}" == "1" ]]; then
if command -v docker >/dev/null 2>&1 && docker compose version >/dev/null 2>&1; then
run "local-prod-infra-lab" bash "$TESTS/phase-prod-infra-lab-tests.sh"
packages/casan-harness/scripts/bash/goal-orchestrator.py
+35 -4
View File
@@ -267,6 +267,16 @@ def extract_patch(text: str) -> str:
return patch
def validate_write_output(text: str) -> str:
"""Fail at the producing harness when a write-intent reply is not a diff.
Keeping this separate from storage avoids creating an artifact before the
reviewer has completed, while ensuring H2/H3 accurately identify a model
that violated the required output contract.
"""
return extract_patch(text)
def validate_and_store_patch(job_path: str, job: dict, patch: str) -> dict:
roots = [str(value).strip("/") for value in job.get("workspace", {}).get("context_roots", [])]
changed = []
@@ -640,6 +650,14 @@ def run(job_path: str) -> int:
if not allowed:
emit(goal_id, "H4-security", "blocked", "Local worker output rejected")
raise ValueError("local_output_security_blocked")
if write_intent:
try:
validate_write_output(safe_local)
except ValueError as error:
reason = str(error)
stage(job_path, "local-worker", "error", reason, job.get("local_provider", ""), local_model)
emit(goal_id, "H2-tool", "error", "Local worker violated patch output contract", {"reason": reason})
raise
stage(job_path, "local-worker", "pass", "Primary solution prepared", job.get("local_provider", ""), local_model)
update_job(job_path, local_draft=safe_local, local_usage=local_meta)
emit(goal_id, "H2-tool", "pass", "Local solution prepared", {"provider": job.get("local_provider", ""), "model": local_model, **local_meta})
@@ -670,10 +688,23 @@ def run(job_path: str) -> int:
if not allowed:
emit(goal_id, "H4-security", "blocked", "Cloud reviewer output rejected")
raise ValueError("cloud_output_security_blocked")
stage(job_path, "cloud-reviewer", "pass", "Independent review incorporated", reviewer_provider, reviewer_model)
emit(goal_id, "H3-eval", "pass", "Independent review incorporated", {"provider": reviewer_provider, "model": reviewer_model, **cloud_meta})
final_status = "completed"
metric_status = "success"
if write_intent:
try:
validate_write_output(safe_result)
except ValueError as error:
cloud_ok = False
cloud_reason = f"reviewer_output_contract_invalid:{error}"
if cloud_ok:
stage(job_path, "cloud-reviewer", "pass", "Independent review incorporated", reviewer_provider, reviewer_model)
emit(goal_id, "H3-eval", "pass", "Independent review incorporated", {"provider": reviewer_provider, "model": reviewer_model, **cloud_meta})
final_status = "completed"
metric_status = "success"
else:
safe_result = safe_local
stage(job_path, "cloud-reviewer", "warning", cloud_reason, reviewer_provider, reviewer_model)
emit(goal_id, "H3-eval", "warning", "Cloud reviewer output contract rejected; local solution retained", {"reason": cloud_reason})
final_status = "degraded"
metric_status = "degraded"
else:
safe_result = safe_local
stage(job_path, "cloud-reviewer", "warning", cloud_reason, reviewer_provider, reviewer_model)