Skip to content

The agent that grows with you

chore(catalog): review disclosure for hermes-guide

Teknium8a4510079b5e
diff --git a/plugin-catalog/hermes-guide.yaml b/plugin-catalog/hermes-guide.yaml
index ecf71b9fa46..4a32c30fb87 100644
--- a/plugin-catalog/hermes-guide.yaml
+++ b/plugin-catalog/hermes-guide.yaml
@@ -1,7 +1,7 @@
name: hermes-guide
repo: https://github.com/iap/hermes-guide
sha: ca10574b22c703215fb5321dd32568d4203ba472
-description: "Read-only diagnostics for Hermes configuration: /hermes-doctor in-session and hermes guide in a terminal. Ten read-only health checks (config, mcp, skills, commands, hooks, plugins, memories, sessions, storage, update) and points to 22 troubleshooting skills, installed separately as a skills tap, for what the built-in doctor doesn't cover. Complements — never replaces — hermes doctor. Disclosure — reads config.yaml, SKILL.md frontmatter, skill bundles, plugin manifests, the shell-hooks allowlist and the built-in MEMORY.md/USER.md stores read-only (the report quotes short memory snippets); the sessions scope reads the tail of logs/agent.log read-only for runaway-session signatures (session ids, model names and token counts appear in the report); the update scope reads the update-lock record in the co-located hermes-agent checkout under $HERMES_HOME and probes it — owner-PID liveness, then a brief lock acquisition it releases immediately, opening the lock file writable, with a read-only fallback for a root-owned lock on POSIX (writes no content, never deletes the lock); registers one read-only plugin skill, hermes-guide:start, listed in model sessions while the plugin is enabled; /hermes-doctor and hermes guide run hermes config path as a subprocess (argument list, no shell); the default hooks check reads hook configuration, consent state and the exec bit in-process and executes nothing, while hooks --deep runs hermes hooks doctor, which executes every approved shell hook once with a synthetic payload (--fast starts no subprocess and executes no hook); /hermes-doctor and hermes guide (except with --fast or --json) also run git ls-remote once per process against the plugin's GitHub repo to print a newer-release notice (no download, no local writes); with opt-in proactive: true it runs the checks (including the agent.log scan and the update-lock probe) in-process at each session start/end (no hooks executed, no subprocesses, no network). The 22 troubleshooting skills are not loaded by the plugin: they install separately from the repo as a skills tap, which this entry's pin does not cover."
+description: "Read-only diagnostics for Hermes configuration: /hermes-doctor in-session and hermes guide in a terminal. Ten read-only health checks (config, mcp, skills, commands, hooks, plugins, memories, sessions, storage, update) and points to 22 troubleshooting skills, installed separately as a skills tap, for what the built-in doctor doesn't cover. Complements — never replaces — hermes doctor. Disclosure — reads config.yaml, SKILL.md frontmatter, skill bundles, plugin manifests, the shell-hooks allowlist and the built-in MEMORY.md/USER.md stores read-only (the report quotes short memory snippets); the sessions scope reads the tail of logs/agent.log read-only for runaway-session signatures (session ids, model names and token counts appear in the report); the storage scope stats $HERMES_HOME/state.db for its size only (never opens it); the update scope reads the update-lock record in the co-located hermes-agent checkout under $HERMES_HOME and probes it — owner-PID liveness, then a brief lock acquisition it releases immediately, opening the lock file writable, with a read-only fallback for a root-owned lock on POSIX (writes no content, never deletes the lock); registers one read-only plugin skill, hermes-guide:start, listed in model sessions while the plugin is enabled; /hermes-doctor and hermes guide run hermes config path as a subprocess (argument list, no shell); the default hooks check reads hook configuration, consent state and the exec bit in-process and executes nothing, while hooks --deep runs hermes hooks doctor, which executes every approved shell hook once with a synthetic payload (--fast starts no subprocess and executes no hook); /hermes-doctor and hermes guide (except with --fast or --json) also run git ls-remote once per process against the plugin's GitHub repo to print a newer-release notice (no download, no local writes); with opt-in proactive: true it runs the checks (including the agent.log scan and the update-lock probe) in-process at each session start/end (no hooks executed, no subprocesses, no network). The 22 troubleshooting skills are not loaded by the plugin: they install separately from the repo as a skills tap, which this entry's pin does not cover."
maintainer: iap
tier: community
category: tools