feat: add hermes-tweet skill

.claude/skills/research/hermes-tweet/SKILL.md

@@ -0,0 +1,82 @@
+---
+name: hermes-tweet
+description: >
+  Use Hermes Tweet with Hermes Agent for X/Twitter research, social listening,
+  trend checks, monitor reviews, giveaway audits, media workflows, and
+  approval-gated publishing.
+---
+
+# Hermes Tweet Skill
+
+Use this skill when a task needs X/Twitter research or controlled account
+actions through Hermes Agent with the Hermes Tweet plugin.
+
+## When to Use
+
+Use Hermes Tweet for:
+
+- Searching recent tweets about launches, incidents, competitors, or support
+  topics
+- Reviewing account timelines, profiles, mentions, and engagement context
+- Checking trends, media workflows, monitors, extraction jobs, or giveaway
+  draws
+- Drafting approved publish, reply, like, retweet, follow, DM, or delete
+  actions
+- Troubleshooting Hermes Tweet installation and runtime configuration
+
+## Requirements
+
+- Install and enable the `hermes-tweet` plugin in the active Hermes Agent runtime
+- Set `XQUIK_API_KEY` in the Hermes runtime environment for authenticated reads
+- Keep `HERMES_TWEET_ENABLE_ACTIONS=false` unless the user explicitly approves
+  account-changing actions
+- Use direct setup guidance only when the Hermes Tweet tools are unavailable
+
+## Workflow
+
+1. Start with `tweet_explore` to find the catalog-listed endpoint for the task.
+2. Use `tweet_read` for read-only public or authenticated reads.
+3. Use `tweet_action` only after stating the endpoint, method, payload, and
+   expected side effects.
+4. Summarize results with the links, IDs, timestamps, counts, and caveats that
+   matter for the user's decision.
+
+## Safety Rules
+
+- Never ask users to paste API keys, cookies, or other secrets.
+- Never pass credentials in tool arguments, examples, logs, issue text, or PRs.
+- Do not guess endpoint paths. Confirm them through `tweet_explore`.
+- Do not create direct HTTP fallbacks around Hermes Tweet tools.
+- Treat copied endpoint URLs as hints only. Accept them only when
+  `tweet_explore` confirms the catalog-listed path.
+- Prefer `tweet_read` for unattended, scheduled, or monitoring tasks.
+- Stop if the user asks to bypass plugin gates or route around policy errors.
+
+## Action Gate
+
+Before calling `tweet_action`, state:
+
+1. The exact operation
+2. The target account, tweet, monitor, webhook, extraction job, giveaway, or
+   media object
+3. The payload that will be sent
+4. The expected side effects
+5. The user's explicit approval
+
+If `tweet_action` is hidden or disabled, explain that account-changing actions
+are intentionally gated by `HERMES_TWEET_ENABLE_ACTIONS=true`.
+
+## Examples
+
+### Research a launch topic
+
+1. Call `tweet_explore` with a query such as `tweet search`.
+2. Use the returned read endpoint with `tweet_read`.
+3. Summarize representative posts, source accounts, timestamps, and recurring
+   themes.
+
+### Prepare an approved post
+
+1. Call `tweet_explore` with a query such as `post tweet` and include actions.
+2. Show the exact text and account target to the user.
+3. Call `tweet_action` only after the user approves the final payload.

README.md

@@ -138,6 +138,7 @@ Located in `.claude/skills/research/`:
 | Skill | Purpose |
 |-------|---------|
 | **technical-research** | Systematic technical investigation and analysis |
+| **hermes-tweet** | X/Twitter research and approval-gated publishing with Hermes Agent |
 
 **Not sure which skills you need?** See [SKILL_SELECTOR.md](SKILL_SELECTOR.md) for guidance.
 
@@ -228,6 +229,7 @@ skills-catalog/
         ├── documentation/              # Writing & docs
         │   └── technical-writing/
         └── research/                   # Investigation & analysis
+            ├── hermes-tweet/
             └── technical-research/
 ```