You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The triage agent runner does not execute the pre_script field from harness configuration. A per-repo triage harness using base composition declares pre_script: scripts/custom-pre-triage.sh, and that script exits 1 — but the triage agent proceeds to run normally.
This was discovered while verifying the base-composition dispatch fix from fullsend-ai#5693 (closing fullsend-ai#5045). The dispatch enumeration now correctly handles base: fields, so the triage harness is found and dispatched. However, the triage runner path does not invoke the pre_script before starting the agent.
Evidence
Harness: .fullsend/harness/triage.yaml with pre_script: scripts/custom-pre-triage.sh
Script: scripts/custom-pre-triage.sh contains echo "hello world" followed by exit 1
Expected: triage aborts due to non-zero pre_script exit
Actual: triage runs to completion and posts a triage comment
The run.go pre_script handling (lines 760-773) may not be reached by the triage agent's execution path, or the triage reusable workflow does not invoke fullsend run in a way that triggers pre_script execution.
Expected behavior
When a triage harness declares pre_script, the runner should execute it before starting the triage agent. If the script exits non-zero, triage should abort (matching code/fix behavior).
Validation criteria
A triage harness with pre_script pointing to a script that exits 1 should cause the triage run to fail/abort
A triage harness with pre_script pointing to a script that exits 0 should allow triage to proceed normally
pre_script from base-composed harnesses should be respected the same as inline declarations
Problem
The triage agent runner does not execute the
pre_scriptfield from harness configuration. A per-repo triage harness using base composition declarespre_script: scripts/custom-pre-triage.sh, and that script exits 1 — but the triage agent proceeds to run normally.This was discovered while verifying the base-composition dispatch fix from fullsend-ai#5693 (closing fullsend-ai#5045). The dispatch enumeration now correctly handles
base:fields, so the triage harness is found and dispatched. However, the triage runner path does not invoke thepre_scriptbefore starting the agent.Evidence
.fullsend/harness/triage.yamlwithpre_script: scripts/custom-pre-triage.shscripts/custom-pre-triage.shcontainsecho "hello world"followed byexit 1Context
reusable-code.ymlandreusable-fix.ymlcall pre-scripts both inline (in the workflow YAML) and viarun.go:760-773. The triage runner path appears to lack equivalentpre_scriptexecution.run.gopre_script handling (lines 760-773) may not be reached by the triage agent's execution path, or the triage reusable workflow does not invokefullsend runin a way that triggers pre_script execution.Expected behavior
When a triage harness declares
pre_script, the runner should execute it before starting the triage agent. If the script exits non-zero, triage should abort (matching code/fix behavior).Validation criteria
pre_scriptpointing to a script that exits 1 should cause the triage run to fail/abortpre_scriptpointing to a script that exits 0 should allow triage to proceed normallypre_scriptfrom base-composed harnesses should be respected the same as inline declarations