core/agent

Snider 8fed5bc6ef docs: add hook improvement tasks for better feedback cycle

6 new issues for Claude Code hook improvements:

- 012: Test output filtering (reduce noise, show failures only)
- 013: Stop verification (verify work complete before stopping)
- 014: Auto-test on edit (async tests after code changes)
- 015: Session context injection (git/issues/CI on startup)
- 016: Silent auto-formatting (suppress formatter output)
- 017: Expose/hide policy (define what to show vs suppress)

Based on Claude Code hooks documentation:
- PostToolUse with suppressOutput for noise reduction
- Stop hooks with agent verification
- Async hooks for background testing
- SessionStart for context injection
- additionalContext for exposing important info

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

2026-02-01 18:57:16 +00:00

2.8 KiB

Raw Blame History

feat(hooks): Define what to expose vs hide in output

Summary

Document and implement a consistent policy for what hook output should be exposed to Claude vs hidden.

Principles

Expose (additionalContext):

Errors that need fixing
Failures that block progress
Decisions that need to be made
Security warnings
Breaking changes

Hide (suppressOutput):

Success confirmations ("formatted", "passed")
Verbose progress output
Repetitive status messages
Debug information
Intermediate results

Output Categories

Always Expose

Category	Example	Reason
Test failures	`FAIL: TestFoo`	Must be fixed
Build errors	`cannot find package`	Blocks progress
Lint errors	`undefined: foo`	Code quality
Security alerts	`HIGH vulnerability`	Critical
Type errors	`type mismatch`	Must be fixed

Always Hide

Category	Example	Reason
Pass confirmations	`PASS: TestFoo`	No action needed
Format success	`Formatted 3 files`	No action needed
Coverage numbers	`coverage: 84.2%`	Unless requested
Timing info	`(12.3s)`	Noise
Progress bars	`[=====> ]`	Noise

Conditional

Category	Show When	Hide When
Warnings	First occurrence	Repeated
Suggestions	Actionable	Informational
Diffs	Small (<10 lines)	Large
Stack traces	Unique error	Repeated

Implementation Pattern

#!/bin/bash
INPUT=$(cat)
OUTPUT=$(echo "$INPUT" | jq -r '.tool_response.output // empty')

# Extract what matters
ERRORS=$(echo "$OUTPUT" | grep -E "FAIL|ERROR|panic" | head -10)
SUMMARY=$(echo "$OUTPUT" | tail -1)

if [ -n "$ERRORS" ]; then
  # Expose failures
  cat << EOF
{
  "hookSpecificOutput": {
    "hookEventName": "PostToolUse",
    "additionalContext": "## Issues Found\n\n$ERRORS\n\n**Summary:** $SUMMARY"
  }
}
EOF
else
  # Hide success
  echo '{"suppressOutput": true}'
fi

Hook-Specific Policies

PostToolUse (Bash)

core go test  → Expose failures, hide passes
core go lint  → Expose errors, hide "no issues"
core build    → Expose errors, hide success
git status    → Always expose (informational)
git commit    → Expose result, hide diff

PostToolUse (Write/Edit)

Format result → Always hide
Debug check   → Expose if found, hide if clean

PostToolUseFailure

All errors    → Always expose with context
Interrupts    → Hide (user initiated)

Aggregation

When multiple issues, aggregate intelligently:

Instead of:
- FAIL: TestA
- FAIL: TestB
- FAIL: TestC
- (47 more)

Show:
"50 tests failed. Top failures:
- TestA: nil pointer
- TestB: timeout
- TestC: assertion failed
Run `core go test -v` for full output"

2.8 KiB Raw Blame History