tres-request-skill-update

TRES Plugin Feedback

Guide the user through submitting clear, actionable feedback about the TRES Claude plugin and save it via the save_ai_conversation_feedback TRES MCP tool. No extra configuration required — works for any user with a TRES API token.

This skill operates in two modes:

1. Explicit feedback — the user asks to submit feedback, report a bug, etc. Run the full guided flow starting at Step 1.
2. End-of-conversation prompt — the conversation is wrapping up. Offer a quick, non-intrusive nudge (see Step 0) and only proceed to the full flow if the user opts in.

---

Step 0 — End-of-Conversation Prompt

When the conversation appears to be ending (the user says "thanks", "that's all", "looks good", etc.) and this skill is triggered proactively:

Ask casually:

Before you go — any feedback on how this session went? Bugs, ideas, or things that could be better? Happy to log it for the team in 30 seconds.

• If the user declines ("no", "I'm good", "nah") — say "No worries, have a great day!" and stop. Do NOT push further.
• If the user shares something — transition into Step 1 and run the full guided flow, but keep it lightweight. Skip questions they've already answered in the conversation and aim for a quick turnaround.
• Keep the tone light and optional — this should never feel like a mandatory survey.

---

Step 1 — Identify Submitter

1. Call get_viewer (TRES MCP, no arguments) to identify the submitter.

   • Extract the organization name from the response.
   • If get_viewer fails, use "Unknown org" as the submitter and continue.

2. Get the user's git identity for the feedback record:

   git config user.name && git config user.email
   

   Use line 1 as agent_name and line 2 as agent_email. If git is not available, use the system username and leave email empty.

Tell the user: "You're connected as {orgName}. What feedback would you like to share?"

---

Step 2 — Understand the Feedback

Let the user describe their feedback in their own words first. Then classify it into one of these categories based on what they said:

Category	When to use
Bug report	Something is broken, erroring, or producing wrong results
Feature request	A new capability that doesn't exist yet
Improvement	An existing feature works but could be better (UX, formatting, flow, performance)
New skill idea	A proposal for an entirely new skill
MCP / data issue	The TRES MCP is returning wrong data, missing fields, or behaving unexpectedly
Positive feedback	Something is working well and the user wants the team to know
General feedback	Anything else — workflow friction, confusion, documentation, onboarding

Confirm the category with the user: "It sounds like this is a {category} — is that right?"

If the feedback is about a specific skill or MCP tool, identify which one. If it's general or about the plugin overall, note that.

Known skills:

Skill	Description
tres-asc845-swap-reprice-skill	ASC 845 swap repricing to zero clearing account residuals
tres-explorer-tx-to-ledger	Add explorer TX to the TRES ledger
tres-tx-story	TX flow diagram and explanation
tres-recon-gaps	Reconciliation gap resolution
tres-asset-balance-validation	Balance validation vs DeBank
tres-report-analyzer	Analyze TRES report XLSX exports
tres-report-advisor	Recommend the right TRES report
tres-invoice-bill-matching	Match txs to ERP invoices/bills
tres-export-3rd-party-contacts	Export unidentified counterparties to XLSX
tres-import-contacts	Import contacts from CSV/XLSX
tres-cost-basis	Cost basis calculation, strategy, issues, reevaluations, exports
tres-rollup-rules	Sub-transaction rollup rules (aggregate txs)
tres-onboarding	Full entity onboarding (orchestrates sub-skills)
tres-settings-management	Org and platform settings
tres-wallets-upload	Wallet onboarding
tres-upload-tx-header-validation	Bulk transaction CSV header naming validation
tres-request-skill-update	This feedback skill

Maintainer note: update this table when new skills are added to the plugin.

---

Step 3 — Dig Deeper

Based on the category, ask targeted follow-up questions to make the feedback actionable. Ask one or two questions at a time — keep it conversational, not interrogative.

For bug reports:

• What exactly happened? What did you see?
• What did you expect to happen instead?
• What were you doing when it happened? (steps to reproduce)
• Did you see an error message? If so, what did it say?
• Can you share a specific example? (tx hash, wallet address, input you used)

For feature requests & improvements:

• What problem would this solve for you?
• How do you handle this today without the feature?
• Can you describe what the ideal experience would look like?
• How often do you run into this need?

For new skill ideas:

• What workflow or task would this skill automate?
• Who on your team would use it, and how often?
• Can you walk through a concrete example of how you'd use it?
• What data source would it need? (TRES MCP, external API, local files)

For MCP / data issues:

• Which MCP tool or query was involved?
• What data did you get back, and what was wrong about it?
• What did you expect the data to look like?
• Can you share the specific query or identifiers you used?

For positive feedback:

• What specifically worked well?
• Was there anything that surprised you (in a good way)?
• Is there a particular workflow or use case where it really shined?
• Would you change anything to make it even better?

For general feedback:

• What part of the experience are you reacting to?
• Was anything confusing or unclear?
• What would have made it better?

Adapt based on what the user has already told you — skip questions they've already answered. The goal is to get enough detail that someone reading the feedback can understand the context and take action without needing to ask follow-up questions.

---

Step 4 — Preview the Feedback

Compose the feedback and show it to the user for review. Format it clearly:

Headline: {concise summary — max 80 chars}

Category: {Bug report | Feature request | Improvement | New skill idea | MCP issue | Positive feedback | General}
Area: {skill name, "MCP", or "General"}
Org: {orgName}

---

{Well-structured description that includes:
 - What the feedback is about (context)
 - The core issue, idea, or praise (substance)
 - Supporting details — steps to reproduce, examples, expected behavior, etc.
 - Impact — how often this comes up, how many people it affects, how it blocks work}

Tags: {comma-separated list}

Writing the description:

• Synthesize the user's answers into a clear, readable narrative — don't just dump Q&A pairs.
• Lead with the most important point.
• Include concrete details (tx hashes, error messages, specific steps) — these are what make feedback actionable.
• For positive feedback, be specific about what worked and why it mattered.

Also show the user the conversation excerpt that will be submitted (see Step 5 format). This gives them informed visibility into what leaves their machine.

Ask: "Here's what I'll submit — does this capture everything? Want to change anything?"

Iterate if the user wants edits. Only proceed on explicit confirmation.

---

Step 5 — Submit Feedback

Call save_ai_conversation_feedback with these arguments:

Argument	Value
headline	Concise summary (max 80 chars). Prefix with category: [Bug], [Feature], [Improvement], [New Skill], [MCP Issue], [Praise], or [Feedback]
description	The full structured description from Step 4
conversation	Scoped and redacted conversation excerpt (see format below)
tags	Array — always include "plugin-feedback", plus the category tag ("bug", "feature", "improvement", "new-skill", "mcp-issue", "praise", "general"), plus the skill name if applicable
agent_name	From git config (Step 1)
agent_email	From git config (Step 1)

Conversation format — include only the last 30 exchanges (user+assistant pairs) from your context window, not the entire session history. Apply a redaction pass before including any message:

• Replace Bearer [A-Za-z0-9._\-]{20,} → Bearer [REDACTED]
• Replace standalone 64-character hex strings → [REDACTED_HEX]
• Replace file paths matching /Users/<name>/ or /home/<name>/ → /Users/[REDACTED]/
• Remove any apparent BIP-39 mnemonic phrases (12–24 dictionary words)

Format:

USER: <text, redacted>

ASSISTANT: <text, redacted>

TOOL_CALL: <tool_name>(arg1=value1, arg2=value2)
TOOL_RESULT: <one-line summary of the result>

... continue for the last 30 exchanges only ...

---

Step 6 — Report Result

On success: Tell the user:

Your feedback has been submitted — thank you! The team will review it.

If the feedback was a bug or blocker, add:

If this is urgent, reach out to the team directly as well.

On failure: Surface the error from the MCP tool and suggest:

• Check that the TRES API token is valid
• Try again — it may be a transient issue

---

Error Handling

Situation	Action
get_viewer fails	Continue with "Unknown org" as submitter
save_ai_conversation_feedback fails	Surface the error; suggest checking TRES API token
User cancels at preview	Say "No problem — feedback was not submitted" and stop
User provides empty description	Ask again — a description is required
git config unavailable	Use system username for agent_name, leave agent_email empty