reconstruct-system-intent

Announce at start: "Using codebase-analyzer to reconstruct system intent."

Overview

What was this system designed to become? Not what it claims to do, but what the architecture, gates, and hidden capabilities reveal about its true purpose. The real moat is rarely in the client binary -- it is in the service/backend layer.

Prerequisite: Reads ALL completed analysis files from docs/analysis/. This is the terminal skill -- it consumes every prior analysis output.

Five Intent Questions

Answer these by synthesizing evidence across all analysis phases:

1. What is this system designed to become? Sources: architecture.md + tool-graph.md + gate-map.md + git archaeology (evolution trajectory) Look for: abandoned features revealing planned direction, investment patterns in service layer vs client, architectural decisions that only make sense at scale.

2. Where is the moat? Sources: gate-map.md (feature gate analysis) + tool-graph.md (tool registration) -- client vs service vs ecosystem Look for: where the most complex logic lives, which capabilities are server-controlled, what would be hardest to replicate.

3. What can it do that it does not expose? Sources: tool-graph.md + gate-map.md + dead-code.md Look for: conditionally registered but currently dormant tools, hidden admin capabilities, API endpoints behind disabled feature flags.

4. How is behavior really controlled? Sources: prompt-influence.md + gate-map.md (feature gates) Look for: remote configuration driving tool availability, prompt instructions shaping behavior at inference time, server-driven capability gating.

5. What are the hidden dependencies? Sources: provenance.md + build-pipeline.md Look for: undocumented telemetry endpoints, data sent to services not mentioned in public docs, build-time pins to specific infrastructure.

Synthesis Output

Produce a single comprehensive report:

# Codebase Analysis Report
## Date: [YYYY-MM-DD]
## Target: [repo path/description]

### Executive Summary
(2-3 sentences: what the system is, its maturity, and the most important finding)

### Track A: Code Quality Findings
- Target classification and tech stack
- Architecture overview
- Dependency health
- Dead code inventory
- API surface summary
- Code quality assessment
- Refactoring recommendations (prioritized by impact/effort)

### Track B: System Intelligence Findings
- Provenance and build pipeline analysis
- Agent loop and tool graph
- Conditional behavior and feature gates
- Prompt influence analysis
- Threat model (what the system could do if misused)

### System Intent Narrative
(Answer each of the five intent questions with evidence and confidence level)

### Confidence-Weighted Evidence Map
| Finding | Confidence | Evidence Sources | Gaps |
|---------|-----------|-----------------|------|
| (finding) | High/Med/Low | (which files/analysis) | (what would strengthen) |

### Priority Actions
1. (highest impact, most urgent)
2. ...