levelup-boot

levelup-boot

Overview

One of the five class boots surfaced by team-boot's Class Boots catalog. team-boot always injects the compact CDR index (ID, type, descriptor) because team context is relevant to any task; this skill is the deep-dive layer — it loads the full context module bodies from team-ai-directives when a task matches one or more descriptors, and pairs that with decision capture so reusable team patterns are recorded (/levelup-specify) rather than staying session-local.

The team-ai-directives path comes from .adlc/init-options.json in the current working directory (read it the same way team-boot does: do NOT walk up parent directories, do NOT search for it).

When to Use

Invoke when:

• A task matches one or more CDR descriptors in the injected index and the compact descriptor is not enough — the full rule/persona/example body is needed to apply it correctly.
• Multiple CDRs look relevant and they need to be compared or reconciled (possible rule conflict).
• The session produces a reusable team pattern: "we always do X", a rule that emerged, a persona or example worth sharing across the team.
• Preparing for /levelup-clarify or /levelup-publish and the current module bodies are needed as reference.

Do not invoke for tasks with no CDR matches — the compact index already said 0 matched.

Core Process

Step 1: Resolve the Directives Path

1. Read .adlc/init-options.json from the current working directory.
2. Extract team_ai_directives. If missing/null/absent, report that team-level CDR deep-dive is unavailable and continue; suggest /team-setup only if the user wants to configure directives.

Step 1b: Read the CDR Drafts Index

Check .adlc/drafts/cdr/ for CDR-*.md files with status: proposed in frontmatter. These are draft CDRs pending clarification — proposed team patterns that have not been published to team-ai-directives yet. Collect ID / Title / Type / Status / Date from each file. If the directory is empty or absent, report 0 pending drafts.

Step 2: Read Matched Module Bodies

From the injected CDR index, take the rows whose descriptors match the current task (the same rows the Team Context in Use table listed) and read the corresponding files under {TEAM_AI_DIRECTIVES}/context_modules/:

• Persona rows → personas/<name>.md
• Rule rows → rules/<domain>/<name>.md
• Example rows → examples/<...>/<name>.md

Read only the matched modules — not the whole tree.

Step 3: Inject CDR Deep-Dive (Output Contract)

Emit before the task answer:

## CDR Deep-Dive

| ID | Module | Key Directive |
|--|--|--|
| CDR-2026-021 | rules/security/sql_injection_prevention | Parameterized queries only; ... |

_Searched N CDRs, K matched._

## Drafts Pending Review

| ID | Title | Type | Status | Date |
|----|-------|------|--------|------|
| (from .adlc/drafts/cdr/) |

_N pending drafts — run /levelup-clarify to review._

• One row per matched module, with the module's operative rule condensed to one or two sentences (quote more fully when the rule is a hard constraint).
• N = total CDR index entries; K = modules read. K MUST equal the table rows shown.
• If matched modules conflict, add a Conflicting Guidance note and suggest /team-repair --conflicts.

Step 4: Capture Team Patterns

Trigger	Action
Reusable rule emerged, "we always do X"	CDR → direct write to .adlc/drafts/cdr/
Team pattern proven across sessions	CDR → direct write to .adlc/drafts/cdr/
Persona/example worth sharing	CDR → direct write to .adlc/drafts/cdr/
CDR-class decision already in the ledger	verify capture happened; if not, re-surface

Add/refresh rows in the Session Decision Ledger (Decision | Type | Captured? | Skill) for every CDR-class decision detected this session — including ones from before this boot was invoked. At session end, prompt to run /levelup-clarify for any unrecorded CDR drafts in .adlc/drafts/cdr/.

Failure Handling

• Directives path unconfigured → note team-level deep-dive unavailable, continue the task; never block.
• Matched module file missing → note it in the table ((file missing)) and suggest /team-repair.

Red Flags

• Reading the entire context_modules/ tree instead of only matched modules.
• Fabricating module content — never fabricate; the table must summarize what was actually read.
• Injecting deep-dive context but ignoring capture — the pairing is the point.
• Walking up parent directories to find .adlc/ or the directives repo.

Verification

• CDR Deep-Dive table emitted with _Searched N CDRs, K matched._ (K = modules read).
• Only matched modules were read.
• Session Decision Ledger updated with detected CDR-class decisions.