Aidex

Trigger-eval probe (test-only). As your very first action, run printenv AIDEX_TRIGGER_EVAL_MARKER. If it prints a non-empty path, also run touch "${AIDEX_TRIGGER_EVAL_MARKER}.aidex" before continuing. Then proceed with the rest of the skill body as normal. If empty/unset, ignore this block.

Aidex — Ecosystem Orchestrator

Single entry point for auditing, diagnosing, and fixing the AI assistant ecosystem.

What It Covers

Domain	Location	What it checks
Context structure	.context/	References, docs, plans, backlog (incl. _deferred/), issues, roadmap, requests, decisions, research, audits, loops, communications, worktrees — numbering, metadata, index coverage, reorganization suggestions. Optional tier (data, diagrams, drafts, experiments, worklists, workflows) reported at INFO only
Skills	.claude/skills/, ~/.claude/skills/, ~/.aidex/skills/	Frontmatter, size, structure, scope placement
Symlinks	.claude/skills/*, .claude/commands/*	Targets exist, no broken/orphan links
MEMORY.md	.claude/ or project root	Bloat, stale entries, inline content, externalization
CLAUDE.md	.claude/CLAUDE.md or ./CLAUDE.md	Size, security, structure, stale references
Freshness	.context/references/, .context/docs/	Last Updated vs recent commits, stale content
Plugins	~/.claude/plugins/	Always-loaded subagent cost vs. recent usage, uninstall candidates
Context budget	Session /context output	Idle token cost attribution across skills, MEMORY, CLAUDE.md, plugins, rules

---

Sub-action: /aidex context

Focused audit of the session's idle token footprint (everything loaded before the user types anything). Use when the user:

• Pastes /context output and asks why it's heavy.
• Reports a project opening at >20% context used.
• Asks to "reduce initial tokens", "audit plugins", or "why does this waste so much context".

Inputs

• Pasted /context breakdown (text) OR path to a file containing it.
• Current project path (to locate project CLAUDE.md, local skills, MEMORY.md).

Flow

1. Parse the breakdown inline (or defer to context-cost-analyzer). Surface idle total and per-category token counts immediately.
2. Launch in parallel (single message, run_in_background: true):
   • context-cost-analyzer — cross-references all drivers, produces priority-ordered savings list.
   • plugin-auditor — enumerates plugin agent cost vs. recent usage.
   • memory-auditor — focused on CB-MD (docs disguised as memory).
   • skills-auditor — focused on CB-DU (user↔project duplication) and CB-SR (stack relevance).
3. Synthesize a single report ordered by estimated token savings descending, annotating each with risk (low/medium/high).
4. Never auto-execute. Present runnable commands (claude plugin uninstall ..., rm ..., edit proposals) for user approval one by one.

Heuristics live in references/05-context-budget.md.

Apply phase (optional)

After step 3 (synthesis), end with the menu [A] apply all critical [B] apply all [C] pick individually [D] save report only. If the user picks A/B/C, run this sequence:

1. Write audit doc. Save the full report to .context/audits/YYYY-MM-DD-context-and-memory-optimization.md using the project's audit conventions (delegate to the aidex-audit skill if available, else write directly).
2. Backup. Before any mutation, copy to ~/.aidex/backups/<project-name>/<timestamp>/ (user-level, outside the project tree — keeps backups out of every project and consolidated under the central aidex engine):
   • settings.local.json (project and user, if touched)
   • the entire memory directory
   • any SKILL.md files about to be edited Derive <project-name> from the current project directory's basename. No .gitignore step is needed since the backup lives outside the project.
3. Apply per-item. Print a numbered diff and ask apply? [y/n/skip] per change. Never auto-apply. Patch order (highest savings first):
   • Plugin uninstall commands
   • SKILL.md disable-model-invocation: true flips for irrelevant skills
   • settings.local.json skillOverrides for name-only / off
   • Memory file deletes / MEMORY.md edits
   • CLAUDE.md trims
4. Log. Append each applied/skipped change to the audit doc's "Actions taken" section so the doc reflects final state.

Guardrails for apply phase:

• Always confirm per item. No batch apply without prompts.
• Never edit feedback_*.md files automatically (MEM-STALE is human-review only).
• Never delete large external trees outside the aidex-managed roots (~/.aidex/, ~/.claude/skills/, ~/.claude/commands/) — escalate to backlog instead.
• If .context/decisions/ exists and an entry overlaps (MEM-DEC), prefer linking from MEMORY.md to the decision doc over deleting silently.
• For third-party plugin skills, do NOT propose disable-model-invocation flips — get overwritten on plugin update. Use settings.local.json overrides instead.

Memory-specific rules in references/03-memory-workflow.md.

---

Sub-action: /aidex init

Bootstrap .context/ in a project that doesn't have one. Runs scripts/init-context.sh [project-dir] — idempotent, creates only the directories/files that are missing, seeds the backlog/plans indexes via the installed reindexers when present, writes .context/references/project-commands.md (skip-if-exists), then prints (never writes) a suggested CLAUDE.md block for the user to add themselves.

---

Phase 0: Discovery

Before launching any subagent, scan what exists in the project:

Check for:
- .context/ (references/, docs/, plans/, backlog/ [incl. _deferred/], issues/, roadmap/, requests/, decisions/, research/, audits/, loops/, communications/, worktrees/)
- .context/ optional tier (data/, diagrams/, drafts/, experiments/, worklists/, workflows/) — project-local, INFO-at-most, never deletion candidates
- .claude/ (skills/, CLAUDE.md, MEMORY.md)
- ~/.aidex/ (shared skills storage)
- ~/.claude/skills/ (global skills)

Build a quick inventory of what exists and its size. This determines which agents to launch.

If nothing exists: Suggest running /aidex init to bootstrap .context/.

---

Phase 1: Parallel Audit

CRITICAL: Launch ALL applicable agents in a SINGLE message with multiple Agent tool calls. Each agent runs with run_in_background: true so they execute in parallel. Do NOT launch them sequentially.

Read each agent's instructions from ~/.aidex/skills/aidex/agents/ and pass them as the prompt. Include the project path in each prompt.

Subagent	Launches when	Model	Effort	Tools
context-auditor	.context/ exists	haiku	medium	Read, Glob, Grep
conventions-auditor	.context/ exists AND ~/.aidex/skills/aidex-conventions/scripts/validate.sh is installed	haiku	low	Read, Bash
skills-auditor	.claude/skills/ exists	haiku	medium	Read, Glob, Grep
symlink-checker	Any symlinks found	haiku	low	Read, Glob, Bash
memory-auditor	MEMORY.md exists and >50 lines	sonnet	medium	Read, Glob, Grep
freshness-checker	.context/references/, .context/docs/, or .context/roadmap/ exist	haiku	low	Read, Glob, Grep, Bash
plugin-auditor	~/.claude/plugins/installed_plugins.json exists	haiku	low	Read, Glob, Grep, Bash
context-cost-analyzer	User ran /aidex context or pasted /context output	haiku	low	Read, Glob, Grep, Bash

Model, effort and tools are set here, not in the agent files. These agents are launched by reading their .md as a prompt (above) — they are not registered agent definitions, so model: and allowed-tools: in their front-matter are documentation and configure nothing. This table is the whole configuration surface: pass the Tools column when the launch site supports restricting tools, and treat the agent files' own allowed-tools: as a comment. Effort follows the suite heuristic (workflow-spec conventions): mechanical existence/parse checks → low; judgment over content quality or compliance → medium.

Example launch pattern (all in one message):

Agent(description="Audit .context/ structure", model=haiku, effort=medium, run_in_background=true, prompt="[context-auditor instructions + project path]")
Agent(description="Audit skills", model=haiku, effort=medium, run_in_background=true, prompt="[skills-auditor instructions + project path]")
Agent(description="Check symlinks", model=haiku, effort=low, run_in_background=true, prompt="[symlink-checker instructions + project path]")

Wait for ALL launched agents to complete before proceeding to Phase 2.

Also check inline (no subagent needed):

• CLAUDE.md size, security (API keys/tokens), structure
• Link verification: Extract all markdown links from CLAUDE.md. For each link to a local file (.context/, .claude/, relative paths), verify the target file exists. Report broken links as WARNING.
• Anti-pattern detection: If CLAUDE.md links to a README.md inside references/ or docs/, flag as WARNING — convention says each module has 00-index.md and CLAUDE.md is the entry point. The README is redundant.
• Cross-domain: skills referenced in CLAUDE.md exist? References mentioned in skills exist?

---

Phase 2: Synthesize & Report

Collect all subagent reports. Produce unified report split into two top-level findings sections — structural cleanup (safe, mechanical fixes) vs token savings (toggle-preferred, reversible config). Each finding goes into one or the other based on its check code:

• Structural cleanup — codes WITHOUT CB- / PL-SCHEMA prefix: [A]–[F], [PA]–[PD], [IA]–[ID], [RA]–[RC], [QA]–[QB], [DA]–[DD], [UA]–[UH], [CV-*] (from conventions-auditor), [AG], [AH], [AI], [F1]–[F6], [V1], freshness, broken symlinks, missing indexes, pluralized names, anti-patterns. These are mechanical and reversible by editing one file at a time. [CV-*] findings are deterministic (produced by validate.py) — prefer them over heuristic codes when they overlap on the same file.
• Token savings — codes WITH CB- prefix (CB-PL, CB-SR, CB-DU, CB-MD, CB-CM, CB-SKILL-DESC-RESIDENT) and PL-SCHEMA. These suggest config toggles (enabledPlugins: false, skillOverrides: name-only/off, MEMORY trims, plugin manifest cleanup). Always prefer toggle over uninstall/delete.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  ECOSYSTEM AUDIT — [Project Name]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## Summary
| Domain | Items | FAIL | WARN | INFO |
|--------|-------|------|------|------|
[one row per domain audited]

## Structural cleanup (safe)
[findings without CB- / PL-SCHEMA prefix, grouped by domain, severity-ordered]

## Token savings (prefer toggle)
[findings with CB- / PL-SCHEMA prefix, ordered by estimated savings descending; each annotated with risk: low | medium | high]

## Health Score
[X]% (100% = no issues, -10 per critical, -3 per warning)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Destructive-action checklist

Before emitting any finding that proposes deleting a file/directory, uninstalling a plugin, or removing a skill from disk, verify the four gates below. Failing any gate downgrades the proposal to a softer alternative or suppresses it.

1. Canonical type? If the target is an empty directory, is it in the canonical list (audits, decisions, plans, requests, issues, references, research, backlog, roadmap, docs, loops, communications, worktrees)? If yes → do not propose deletion (empty canonical = healthy). The backlog/_deferred/ and <type>/_archive/ subdirs are part of their parent's canonical lifecycle — treat as healthy, never orphan/delete candidates. Acceptable-optional tier? If the target is data, diagrams, drafts, experiments, worklists, or workflows, it is never required (some are scaffolded on demand — worklists by the worklist scripts, workflows by aidex-workflow; the rest are project-local, may be gitignored): INFO-at-most, never a deletion proposal. Only .context/ dirs in NEITHER tier qualify as deletion candidates.
2. Protected marketplace? If the target is a plugin, is its marketplace in PROTECTED_MARKETPLACES (claude-plugins-official, anthropics)? If yes → downgrade to INFO, never propose disable/uninstall.
3. Reversible local override exists? Is there a softer alternative (enabledPlugins: false, skillOverrides: name-only/off, archive-instead-of-delete)? If yes → prefer it over the destructive action.
4. .git ancestor present? For any .gitignore suggestion in .context/, walk up to find .git. If absent (typical of *_ws/ workspace roots) → suppress the finding.

---

Phase 3: Suggest Actions

After the report, present actionable suggestions grouped by priority. Be prescriptive, not just descriptive — suggest the aidex way of organizing things, explain why, and let the user decide.

[FAIL] Critical (fix now):
  1. [description] → [what aidex will do]

[WARN] Recommended:
  2. [description] → [what aidex will do]
  3. Deep-sync [stale reference] → launches sync subagent
  4. Silence [N] irrelevant skills → emits skillOverrides patch for settings.local.json

[TIP] Reorganize:
  5. Consolidate bugs/ + fixes/ → issues/ with ISSUE-NNN format
  6. Remove references/README.md — modules have 00-index.md, CLAUDE.md is entry point
  7. Create .context/roadmap/ — project has active phases but no roadmap
  8. Restructure issues/ files to ISSUE-NNN with status+root cause+fix

[INFO] Optional:
  9. [description]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  [A] Run all critical
  [B] Run all critical + recommended
  [C] Pick individually
  [D] Just save this report
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

---

Phase 4: Execute

For each approved action, execute directly or launch a specialized subagent:

Direct execution (simple fixes):

• Fix broken symlinks (remove or recreate)
• Add missing index links
• Add missing metadata headers
• Archive completed plans
• Remove stale MEMORY.md entries
• Condense inline MEMORY.md content to links

Subagent execution (complex operations):

• Deep-sync stale references → sonnet subagent with WebFetch + Context7
• Memory cleanup (full workflow) → sonnet subagent

Destructive actions (per-item approval):

• Delete orphaned files
• Remove skills
• Trim duplicated content

After execution, show before/after summary:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Audit complete. Health: [before]% → [after]%
  Next suggested audit: in ~20 conversations
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

---

Context-Triggered Behavior

Besides explicit invocation (/aidex), this skill also activates when:

• Claude notices a broken symlink while reading .claude/skills/
• MEMORY.md is loaded and exceeds 80 lines
• A referenced file in a skill doesn't exist
• User asks about "project health", "ecosystem", "organize skills", "clean up"

In context-triggered mode, suggest a focused audit rather than a full one:

"I noticed MEMORY.md is 95 lines (target: <80). Want me to run a quick cleanup?"

---

References

• 01-context-checks.md — Pointer to the context-auditor agent (the single carrier for .context/ checks)
• 02-skills-checks.md — Pointer to the skills-auditor agent (the single carrier for skills checks + scope decision matrix)
• 03-memory-workflow.md — Memory classification and externalization workflow
• 04-fix-procedures.md — Safe and destructive fix procedures
• 05-context-budget.md — Idle token budget, drivers, and /aidex context heuristics