The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted session transcripts and generated files which may contain malicious instructions or prompt injection payloads.
Ingestion points: The skill reads session JSON data and generated repository files through the --session and --files-manifest arguments in SKILL.md and scripts/prepare-share-bundle.mjs.
Boundary markers: references/agentic-setup.md establishes an explicit 'Untrusted content boundary', instructing the agent to treat session content as evidence to be sanitized rather than instructions to be executed.
Capability inventory: The skill utilizes the GitHub CLI (gh) and Node.js scripts to create public branches, upload artifacts, and file issues in external repositories.
Sanitization: scripts/prepare-share-bundle.mjs implements extensive regex-based redaction for PII, secrets, and dangerous paths, supplemented by a Shannon entropy check for high-entropy secret detection and a mandatory manual semantic review gate in references/consent-and-review.md.
[COMMAND_EXECUTION]: The skill executes local system commands to prepare and publish the session bundle.
Evidence: scripts/prepare-share-bundle.mjs invokes git for archive creation. scripts/export-codex-session.mjs spawns the codex app-server to retrieve session data. references/trackers/github-session-share.md uses gh, jq, and base64 for GitHub API interactions. All inputs such as share names and repository handles are validated against strict regex patterns before execution.
[DATA_EXFILTRATION]: The skill is designed to transmit session data to public GitHub repositories.
Evidence: The workflow involves creating a public branch and a feedback issue in a user-configured repository (--issue-repo, --storage-repo). This intentional data sharing is protected by a 'Fresh Informed Consent' gate that requires the user to explicitly acknowledge the public destination and verify artifact hashes before any remote write operation occurs.

om-share-this-session