feat: expose ponytail as MCP prompts, add missing CLI subcommands

[getappz]

Summary

Found via dogfooding: /ponytail* stopped working as Claude Code slash commands once the DietrichGebert/ponytail marketplace plugin (the only thing registering them) was disabled in favor of agentflare's own hook-based routing. Claude Code validates slash commands against its registry before a prompt ever reaches a hook, so disabling the plugin silently broke /ponytail-review, /ponytail-gain, etc. with "Unknown command" errors.

Fixed the way lean-ctx does it for its own /lean-ctx* commands: register MCP prompts (rmcp::model::Prompt/GetPromptResult — same crate agentflare already depends on) instead of relying on a plugin or static command files.

• Adds src/mcp_prompts.rs: 8 prompts (ponytail + its 7 sub-skills: review/audit/debt/gain/help/playbook/no-hallucination).
• Wires list_prompts/get_prompt into AgentflareMcp's ServerHandler impl, enables the prompts capability.
• ponytail (bare) takes an optional mode argument: empty/status reports current mode, lite|full|ultra switches and returns the mode's skill body, off clears it, anything else reports an error — mirroring the original hook's SwitchAction semantics.
• Each sub-skill prompt persists the mode switch (ponytail::set_active) then returns the full skill markdown, matching what the dedicated /ponytail-review etc. commands did originally.
• Adds the two ponytail CLI subcommands (playbook, no-hallucination) that existed in the crate's sub_skills/switcher modules but were never wired to agentflare ponytail <name>, for parity with review/audit/debt/gain.

Test plan

• cargo build --workspace, cargo test --workspace (195 passed, +5 new)
• cargo clippy --workspace --all-targets — no new findings
• End-to-end over raw MCP JSON-RPC: initialize reports the prompts capability, prompts/list returns all 8, prompts/get correctly reports/switches mode and returns full skill content for both the bare ponytail prompt and a sub-skill (ponytail-review)
• After merge, restart Claude Code with agentflare mcp registered and confirm /ponytail* appear in the slash-command autocomplete

Summary by CodeRabbit

• New Features
  • Added new CLI actions to access additional built-in skill content.
  • Introduced MCP prompt support, enabling listing of available prompts and retrieval by name (main mode and per-skill prompts).
  • Added improved handling for mode selection, including “off” and unknown mode requests.
• Bug Fixes
  • Enhanced responses when invalid prompt or mode requests are made, providing more consistent error behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 08c3821b-0429-4e92-809f-5e388fd55b11

📥 Commits

Reviewing files that changed from the base of the PR and between 7468904 and f707597.

📒 Files selected for processing (1)

• src/mcp_prompts.rs

🚧 Files skipped from review as they are similar to previous changes (1)

• src/mcp_prompts.rs

---

📝 Walkthrough

Walkthrough

Adds MCP prompt listing and retrieval for ponytail modes and sub-skills, wires prompt support into the MCP server, and adds CLI actions for two new ponytail sub-skills.

Changes

MCP Prompts feature

Layer / File(s)	Summary
Prompt data and module setup src/main.rs, src/mcp_prompts.rs	Registers the new mcp_prompts module and defines the ponytail sub-skill mapping.
Prompt listing and dispatch src/mcp_prompts.rs	Creates the base ponytail prompt, adds ponytail-{skill} prompts, and routes incoming prompt names to the right handler.
Mode and skill prompt handlers with tests src/mcp_prompts.rs	Implements prompt formatting, mode/status handling, skill lookup, and unit tests for prompt responses.
MCP server wiring src/mcp_server.rs	Enables prompt capability and exposes list/get prompt handlers that delegate to mcp_prompts.
CLI subcommands for new sub-skills src/cli/ponytail.rs	Adds Playbook and NoHallucination CLI actions and prints their sub-skill text.

Estimated code review effort: 2 (Simple) | ~15 minutes

Sequence Diagram(s)

sequenceDiagram
  participant MCPClient
  participant AgentflareMcp
  participant mcp_prompts
  MCPClient->>AgentflareMcp: list_prompts / get_prompt
  AgentflareMcp->>mcp_prompts: delegate request
  mcp_prompts-->>AgentflareMcp: Prompt result or None
  AgentflareMcp-->>MCPClient: ListPromptsResult / GetPromptResult / invalid_params

Loading

Possibly related PRs

• getappz/agentflare#85: Adds the playbook sub-skill and mode wiring that this PR exposes through PonytailAction::Playbook and ponytail::sub_skills::SKILL_PLAYBOOK.
• getappz/agentflare#96: Adds the no-hallucination sub-skill and mode wiring that this PR exposes through PonytailAction::NoHallucination and ponytail::sub_skills::SKILL_NO_HALLUCINATION.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change by mentioning MCP prompts for ponytail and the added CLI subcommands.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/dogfooding-issues

Warning

Tools execution failed with the following error:

Failed to run tools: 13 INTERNAL: Received RST_STREAM with code 2 (Internal server error)

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (2)

src/mcp_prompts.rs (1)

12-20: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add drift-protection test between SUB_SKILLS and ponytail::sub_skills::get.

SUB_SKILLS (names+descriptions) and ponytail::sub_skills::get's match arms must stay in sync, or get_ponytail_skill's unwrap_or_default() silently returns an empty body for a listed prompt. Consider a test asserting ponytail::sub_skills::get(name).is_some() for every SUB_SKILLS entry, plus coverage for the "off" and successful lite/full/ultra switch paths in get_ponytail_mode.

Also applies to: 85-89, 91-135

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/mcp_prompts.rs` around lines 12 - 20, Add a drift-protection test to keep
SUB_SKILLS and ponytail::sub_skills::get in sync: for every entry in SUB_SKILLS,
assert that get(name) returns Some so get_ponytail_skill never falls back to an
empty body via unwrap_or_default(). Also add coverage around get_ponytail_mode
for the “off” case and the successful lite/full/ultra switch paths so the
prompt-mode mapping stays locked to the documented modes.

src/mcp_server.rs (1)

314-333: 🚀 Performance & Scalability | 🔵 Trivial | 💤 Low value

Blocking file I/O inside async handlers.

get_prompt/list_prompts delegate to mcp_prompts, which does synchronous std::fs reads/writes (via ponytail::active_mode/set_active) directly on the async task. Given the small scope (local dev tool, single flag file), impact is limited, but wrapping the sync calls in tokio::task::spawn_blocking would be more idiomatic if this server ever serves concurrent requests.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/mcp_server.rs` around lines 314 - 333, The async handlers list_prompts
and get_prompt in mcp_server.rs are calling mcp_prompts functions that do
synchronous file I/O on the async task. Move the blocking work behind
tokio::task::spawn_blocking (or equivalent) and await the result inside
list_prompts and get_prompt so the RequestContext<RoleServer> handlers stay
non-blocking. Keep the existing ListPromptsResult::with_all_items and
ErrorData::invalid_params behavior, just wrap the mcp_prompts::list_prompts and
mcp_prompts::get_prompt calls safely.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/mcp_prompts.rs`:
- Around line 74-78: The persistence error from ponytail::set_active is being
ignored, so the mode-switch response can claim success even when the active flag
write fails. Update the mode handling in the relevant match arms in
mcp_prompts::assistant_text flow to handle the Result from ponytail::set_active
explicitly instead of calling .ok(), and surface the I/O failure to the caller
by returning an error/failed response or otherwise preventing the success
message when persistence does not succeed.

---

Nitpick comments:
In `@src/mcp_prompts.rs`:
- Around line 12-20: Add a drift-protection test to keep SUB_SKILLS and
ponytail::sub_skills::get in sync: for every entry in SUB_SKILLS, assert that
get(name) returns Some so get_ponytail_skill never falls back to an empty body
via unwrap_or_default(). Also add coverage around get_ponytail_mode for the
“off” case and the successful lite/full/ultra switch paths so the prompt-mode
mapping stays locked to the documented modes.

In `@src/mcp_server.rs`:
- Around line 314-333: The async handlers list_prompts and get_prompt in
mcp_server.rs are calling mcp_prompts functions that do synchronous file I/O on
the async task. Move the blocking work behind tokio::task::spawn_blocking (or
equivalent) and await the result inside list_prompts and get_prompt so the
RequestContext<RoleServer> handlers stay non-blocking. Keep the existing
ListPromptsResult::with_all_items and ErrorData::invalid_params behavior, just
wrap the mcp_prompts::list_prompts and mcp_prompts::get_prompt calls safely.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 77f888db-d144-4a9d-80b9-0ad1bb7087b6

📥 Commits

Reviewing files that changed from the base of the PR and between 4e249fe and 7468904.

📒 Files selected for processing (4)

• src/cli/ponytail.rs
• src/main.rs
• src/mcp_prompts.rs
• src/mcp_server.rs