Tech Debt Tracker

Track technical debt from conversation and PR review comments. Debt items live in projects/commitments/tech-debt/ and are resurfaced weekly.

Mode A: Passive detection

When the user says something implying tech debt ("this is a hack but it works", "we should really refactor the auth module", "adding another TODO"), silently extract it.

Action:

1. Check for duplicates via memory_search in projects/commitments/tech-debt/.
2. Write to projects/commitments/tech-debt/<slug>.md:

---
type: tech-debt
detected_at: <today YYYY-MM-DD>
repo: <current repo context if known, else null>
severity: <high|medium|low>
category: <refactor|performance|security|testing|documentation|architecture>
source: conversation
source_pr: null
---
# <Brief title>
<What the debt is and why it exists.>

## Context
<What prompted the shortcut — deadline, complexity, missing knowledge.>

## Proposed fix
<What a proper fix would look like, if discussed.>

Severity rules:

• high: security shortcuts, data integrity risks, architectural violations
• medium: code quality, missing tests, poor abstractions
• low: style issues, minor workarounds, documentation gaps

3. At a natural pause: "Tracked tech debt: ."

Do NOT interrupt conversation flow.

Mode B: PR review scan (used by triage mission)

When the triage mission processes recently merged PRs, it should scan review comments for tech-debt patterns:

• "can be addressed in follow-up work"
• "not blocking but should fix later"
• "TODO: address in next PR"
• "leaving for now", "good enough for now"
• "nit: ... (not blocking)"
• "tech debt", "hack", "workaround" in review comments

For each match, create a tech-debt item with source: pr-review and source_pr: owner/repo#123.

Mode C: Explicit capture

User says: "add tech debt: the caching layer needs TTL eviction."

Write directly to projects/commitments/tech-debt/, confirm briefly.

Mode D: Listing

User says: "show tech debt" or "debt backlog"

Action:

1. memory_tree("projects/commitments/tech-debt/", depth=1) — list files (skip README.md)
2. memory_read each for title, severity, detected_at, repo, source
3. Display grouped by severity, then age:

## Tech Debt Backlog

### High Severity
- **<title>** (repo: <repo>, <N> days old, source: <conversation|PR #X>) — <category>

### Medium Severity
- ...

### Low Severity
- ...

---
<count> total. <count> chronic (30+ days).
Say "resolve <title>" to mark fixed, or "promote <title>" to create a commitment.
For large items, use `/plan <description>` to create a structured fix plan.

Mode E: Resolution

User says: "resolved the caching TTL debt"

Move to projects/commitments/resolved/ with original type preserved. Confirm.

Mode F: Promotion to commitment

User says: "let's fix the auth refactor" or "promote the caching debt"

Create commitment in projects/commitments/open/ with tags: [tech-debt] and resolution_path: agent_can_handle if it's a code task. For complex items, suggest: "This looks like a multi-step refactor. Use /plan <description> to create a structured fix plan."

Filename conventions

Slugify: lowercase, hyphens, max 50 chars. Prefix with repo slug if known:

• "Auth module needs refactor" in nearai/ironclaw → nearai-ironclaw-auth-refactor.md
• Generic debt → caching-ttl-eviction.md