Claude Code Hook Development

The material emphasizes input validation, path safety, quoting, timeouts, and user confirmation, and warns against logging sensitive information. However, examples for Slack, databases, metrics, and audit logging transmit or persist tool inputs, commands, user data, or environment data without consent, least-privilege, redaction, access-control, or rollback guidance. validate-write.sh also interpolates an unescaped path into JSON, risking malformed output or injection. Points are deducted for incomplete data-flow transparency, isolation, and recovery.

The skill covers configuration, events, inputs, exit codes, and debugging, with several example scripts. However, it first says plugin hooks.json requires a hooks wrapper, while a later plugin configuration example uses top-level events, creating a key configuration contradiction. The prompt-hook supported-event list conflicts with the PostToolUse prompt example. Several referenced validation and test scripts are not included in the supplied material. Examples also use fragile string matching, omit symlink and path-boundary handling, and the linter may terminate early under set -e because of ((warnings++)). Static evidence is insufficient, so the score is conservative.

The audience, trigger phrases, major events, and command/prompt scenarios are reasonably clear. Boundaries, non-fit cases, dependency installation, permission prerequisites, and Claude Code version differences are not systematically stated. The material is primarily English and Bash/Unix-oriented, with no Chinese guidance or examples; it relies on Claude Code, bash, jq, and potentially overseas model/services, without explaining mainland-China reachability or fallback behavior.

The skill has a name, description, version, layered sections, examples, references, and a debugging workflow. Repository material supplies Apache-2.0 licensing, official maintenance provenance, and installation guidance. Points are deducted because there is no changelog, compatibility matrix, dependency-installation guidance, explicit support/update channel, or complete troubleshooting coverage. Several referenced references, examples, and utility scripts are not fully verifiable from the supplied files, and some configuration examples are inconsistent.

It provides directly adaptable configurations and script patterns for common PreToolUse, Stop, and SessionStart hooks, so the core task is plausible on the happy path. However, prompt-hook decision reliability, parallel-execution semantics, output formats, and permission behavior lack observed results; the security examples can produce false positives or false negatives, and some scripts should not be treated as production-ready without review. Static calibration caps effectiveness at 7.

The material includes concrete configurations, scripts, test commands, a linter, and auditable repository CI and license evidence, providing limited reproducibility. No scripts were executed, and there are no real test outputs, coverage results, CI tests covering this skill's key paths, or independent corroborating sources. The listed CI mainly concerns repository plugins and MCP URL liveness, not this skill's behavior. Static calibration caps verifiability at 5.