| name | roadmap-synthesis |
| description | Produces ROADMAP.md from a full-project audit via a project-derived lens set. Covers paradigm detection (deterministic/agentic/hybrid), lens-set derivation (SPIRIT, DORA, SPACE, FAIR, CNCF Platform Maturity, Custom), parallel audit fan-out, lens synthesis, multi-angle reframing, grounded claim generation. Triggers: ultra-in-depth project analysis, spring cleaning roadmap, project state of the union, lens-based audit, strengths/weaknesses/deprecations roadmap, agentic-era project evaluation, AGENTS.md-aware audit, SDLC health audit. Contrast with `roadmap-planning` (prioritizes existing backlog).
|
| allowed-tools | ["Read","Write","Edit","Glob","Grep","Bash"] |
| compatibility | Claude Code |
Roadmap Synthesis
Produce a ROADMAP.md for any project by deriving a project-specific evaluation lens set, running an ultra-in-depth audit through those lenses, and reconciling parallel findings into a grounded, actionable plan. Paired with roadmap-planning, which owns the prioritization and sequencing step.
This skill is loaded by the roadmap-cartographer agent via its skills: frontmatter. It is the audit-to-synthesis half of the roadmap capability; prioritization logic stays in the paired roadmap-planning skill — the two never absorb one another. The lens framework is derived per project from project values + domain constraints + best-fit exemplar, not hardcoded to a universal list.
Satellite files (loaded on-demand):
- references/lens-framework.md -- the 4-step lens-derivation methodology, exemplar lens sets catalog (SPIRIT, DORA, SPACE, FAIR, CNCF Platform Maturity, Custom), lens schema, and SPIRIT worked example
- references/audit-methodology.md -- parallel deep-dive pattern, lens selection rubric, fragment reconciliation
- references/paradigm-detection.md -- deterministic / agentic / hybrid classification heuristics (feeds lens derivation Step 2)
- references/grounding-protocol.md -- every quantitative claim cites a source; verification checklist
- assets/ROADMAP_TEMPLATE.md -- the 10-section ROADMAP.md scaffold (Executive Summary → Weaknesses → Opportunities (forward lines) → Improvement Roadmap → Decision Log)
- assets/audit-fragment-template.md -- per-lens researcher fragment schema (
AUDIT_<lens>.md)
What this skill does
Given any project — Praxion itself or any codebase Praxion helps develop — produce a ROADMAP.md at the project root that captures strengths, weaknesses, improvement opportunities, deprecations, and sequenced work. The skill is paradigm-agnostic (deterministic / agentic / hybrid) and lens-agnostic: it derives the right evaluation lens set for the target project from the project's own values and constraints, rather than applying a fixed universal list.
When to use this skill vs roadmap-planning
Use roadmap-synthesis when… | Use roadmap-planning when… |
|---|
Producing a fresh ROADMAP.md from a full-project audit | Ordering an existing set of ideas (e.g., idea_ledgers/IDEA_LEDGER_*.md) |
| Input is the repository itself (filesystem, CI, docs) | Input is a pre-existing list of candidates with impact/effort fields |
| User asks "what should we build next" without a backlog | User asks "which of these should we do first" |
| Phrases: "spring cleaning", "state of the project", "ultra-in-depth audit" | Phrases: "prioritize backlog", "RICE/MoSCoW", "sequence releases" |
| Lens-based project evaluation is required | Prioritization framework is required |
Output: ROADMAP.md (project root, living) | Output: ROADMAP.md + BACKLOG.md (.ai-work/<slug>/) |
The two skills compose: roadmap-synthesis runs Phases 1–5 and 7; it delegates Phase 6 (rank and sequence) to roadmap-planning's framework selector and Now/Next/Later format.
Pipeline position
Standalone capability — not inserted into the promethean → researcher → systems-architect chain. Activated by:
/roadmap command with optional mode argument (fresh / diff / <focus-area>)
- Explicit
@roadmap-cartographer delegation
- Main agent recognizing roadmap intent (trigger phrases above)
The roadmap-cartographer agent owns the end-to-end workflow and loads this skill plus roadmap-planning via its skills: frontmatter.
Core workflow
Seven phases. The cartographer mirrors this structure; procedural depth lives in the linked references, not here.
- Scope, Paradigm & Lens Derivation — classify target as deterministic / agentic / hybrid → paradigm-detection.md, then derive the project-specific lens set (4-8 lenses drawn from project values + domain constraints + exemplar lens sets) → lens-framework.md. Surface paradigm and proposed lens set to the user (Gate 1).
- Ecosystem Inventory — filesystem scan of skills, agents, rules, commands, hooks, tests, CI,
AGENTS.md/CLAUDE.md, .ai-state/DESIGN.md, .ai-state/decisions/, memory; read the latest .ai-state/sentinel_reports/SENTINEL_REPORT_*.md if present; detect an existing ROADMAP.md for diff mode.
- Parallel Audit Fan-out — spawn N researcher agents (N = lens_count, capped at 6) in parallel, one per lens from the derived set; each writes an
AUDIT_<lens>.md fragment via the fragment template → audit-methodology.md.
- Lens Synthesis — reduce fragments into draft roadmap sections through the derived lens set → lens-framework.md.
- Multi-Angle Reframe — for each of the top 3 weaknesses, articulate ≥2 framings; record the runners-up as Considered Angles. Universal step regardless of lens set.
- Prioritize & Sequence — delegate to
roadmap-planning for framework selection and Now/Next/Later formatting (Gate 2 on proposed deprecations).
- Self-Verify & Emit — grounding check on every quantitative claim → grounding-protocol.md; Gate 3 on phase ordering; write
ROADMAP.md preserving the Decision Log (never rewritten; append-only in diff mode).
Lens framework summary
Lens sets are derived per project, not fixed. Four exemplar sets plus a Custom branch cover the common project classes; the cartographer composes 4-8 lenses from the best-fit exemplar + the project's own values.
| Exemplar | When to use | Lenses |
|---|
| SPIRIT (Praxion exemplar) | Multi-agent dev tools, LLM-app frameworks | Automation · Coordinator Awareness · Quality · Evolution · Pragmatism · Curiosity & Imagination |
| DORA | Continuous-delivery products, SaaS | Deploy Frequency · Lead Time · Change Fail Rate · MTTR |
| SPACE | Developer productivity, internal platforms | Satisfaction · Performance · Activity · Communication · Efficiency (pick 3+) |
| FAIR | Research code, scientific data repos | Findable · Accessible · Interoperable · Reusable |
| CNCF Platform Maturity | Infra platforms, IDPs, K8s operators | Investment · Adoption · Interfaces · Operations · Measurement |
| Custom / Derived | Anything not matching an exemplar | 4-8 lenses composed from project values + universal Quality + universal Docs |
Full methodology (4-step derivation), lens schema (name / definition / sub-questions / evidence / failure signals), and the SPIRIT worked example in lens-framework.md. Lenses are evaluation tools applied to the target project, not the project's own principles — those live at their canonical homes (e.g., Praxion's README.md#guiding-principles) and are not re-authored by the cartographer.
Paradigm detection (summary)
The cartographer classifies the target project from three signal families — dependency manifests (e.g., anthropic, @modelcontextprotocol/*), filesystem markers (agents/, skills/, .claude-plugin/), and directory layout conventions. The classification feeds lens derivation Step 2 (domain constraints) and selects which sub-question set applies within each chosen lens. A misclassification is a named anti-pattern (R15). Full rubric: paradigm-detection.md.
Grounding rule
Every quantitative claim in the emitted ROADMAP.md must cite a source. Counts, percentages, ratios, and dates are rejected by the Phase 7 self-verification pass if unsourced. Qualitative claims prefer evidence where it exists. Accepted citation formats (file references, command output, ADRs, memory entries, sentinel reports, external URLs with fetch dates) and the full verification checklist: grounding-protocol.md.
Multi-angle reframing (Phase 5 inline procedure)
For each of the top 3 weaknesses from Phase 4, the cartographer produces at least two framings before selecting one, then records the rejected framings as Considered Angles in the output.
- State the weakness in the terms that emerged from Phase 4 (the default framing).
- Invert the frame — ask whether the weakness is a symptom of a deeper structural pattern, or whether it is itself a masked strength (e.g., "missing CI integration test" may be "absence of expensive integration flakiness").
- Recombine frames — consider whether two weaknesses share a root cause and collapse into one remediation.
- Select the framing whose remediation is smallest yet addresses the root cause; keep the runners-up as Considered Angles in the roadmap entry.
- Record rejected framings — the ROADMAP template has a
## Considered Angles sub-block under each top-3 weakness. Never drop rejected framings silently; they are the audit's Curiosity trail.
This procedure operationalizes the Curiosity lens (called Curiosity & Imagination in the SPIRIT exemplar) and mitigates R9 (single-perspective bias). It runs regardless of the derived lens set — multi-angle framing is universal.
Composition with roadmap-planning
At Phase 6, the cartographer activates the roadmap-planning skill and delegates:
- Framework selection —
roadmap-planning's decision table picks RICE / MoSCoW / WSJF / Kano / ICE / simple rank based on project context.
- Dependency mapping —
roadmap-planning's dependency graph construction and critical-path identification.
- Now / Next / Later formatting — the default roadmap format per
roadmap-planning for projects with evolving scope.
The cartographer does not re-implement any of this; it hands the synthesized candidates to roadmap-planning and receives a sequenced Now/Next/Later structure back.
Output
The canonical output is ROADMAP.md at the project root — a living document preserved across runs. The cartographer also writes these working files under .ai-work/<task-slug>/:
ROADMAP_DRAFT.md — intermediate draft before Gate 3
AUDIT_<lens>.md — one per researcher (Phase 3 fragments)
PROGRESS.md — phase-transition signals
Use assets/ROADMAP_TEMPLATE.md as the scaffold. The template's 10-section shape (Executive Summary → Decision Log, including the Opportunities (Forward Lines) section between Weaknesses and Improvement Roadmap) is the output contract. In diff mode the cartographer updates sections in place and appends a new Decision Log entry — it never rewrites the Decision Log. Opportunities catalogue forward-looking items (new capabilities, strategic bets, evolution trends) that may or may not be promoted to Improvement Roadmap items in this cycle — cataloguing them keeps the road ahead visible without forcing premature commitment.
Anti-patterns
Distilled from pre-design research and validated by context-engineer review. Each row states the risk and the design's mitigation. Inline list because a reader scanning the skill needs to see them without a second click.
| ID | Anti-pattern | Mitigation |
|---|
| R1 | Hallucinated features or metrics | Grounding protocol + Phase 7 self-verify (unsourced quantitatives rejected) |
| R2 | Roadmap rot | "Last audited" header + diff mode + Decision Log preserved |
| R3 | Premature prioritization before synthesis | Phase 4 synthesis strictly precedes Phase 6 ranking |
| R4 | Cargo-cult methodology (lens set OR prioritization) | Lens derivation (project values + exemplars) + roadmap-planning framework selector |
| R5 | Score worship over judgment | Template Judgment override field per item |
| R6 | Kitchen-sink roadmap | Now cap ≤ 5 items + dedicated Deprecation section |
| R7 | Weaknesses without strengths | Template Section 2 "What's Working" precedes weaknesses |
| R8 | Implementation prescription in roadmap | "Next pipeline action" names downstream agent, not code |
| R9 | Single-perspective bias | N ≥ 3 parallel lenses + Phase 5 multi-angle reframe |
| R10 | Missing user touchpoints | Three mandatory AskUserQuestion gates (scope, deprecations, ordering) |
| R11 | Bus factor | Optional Ownership field in template |
| R12 | Stalled Now items | diff mode surfaces no-progress items |
| R13 | Cookie licking | Optional "next-step by" timestamp |
| R14 | Over-metricization | Template metric cap per section |
| R15 | Paradigm mismatch | paradigm-detection.md + Phase 1 gate |
| R16 | Regeneration destroys institutional memory | Decision Log preserved; diff mode asks before destroy |
| R17 | Skill-agent activation collision with roadmap-planning | Disambiguation clause in both descriptions (this skill and agent) |
Quick reference
| Dimension | Reference |
|---|
| Core commands | /roadmap (fresh | diff | <focus-area>) |
| Default mode | fresh — full audit, writes new ROADMAP.md |
| Incremental mode | diff — re-audits, preserves Decision Log, surfaces deltas |
| Focused mode | <area> — narrows audit weighting to a focus area; still produces full structure |
| Typical runtime (cartographer) | 3–6 parallel researchers + synthesis ≈ minutes, not hours |
| Paired skill | roadmap-planning — activated at Phase 6 |
| Output location | ROADMAP.md at project root |
| Common pitfall | Skipping paradigm detection or lens derivation → R4/R15 false findings (e.g., cargo-culting SPIRIT to a Python library) |
Related Skills
- roadmap-planning -- prioritization and sequencing; activated at Phase 6 of this skill's workflow
- project-exploration -- codebase orientation; useful before audit fan-out when the target project is unfamiliar