Scan

Quick Summary

Goal: Scan the codebase for ONE target reference doc and surgically (re)populate it with actual patterns — every example from real project files with file:line. The 4-phase engine below is shared; the per-target data (which doc, how many sub-agents, what to detect, what sections to write, what NOT to do) comes from the target's entry in references/targets.md.

Workflow:

1. Resolve target — Read --target=<key>; load its entry from references/targets.md
2. Assess — Read target doc, detect init vs sync (vs force) mode, run the target's Phase-0 detection table(s)
3. Scan — Launch the target's sub-agents in parallel; discover patterns with file:line evidence
4. Report — Write structured findings to report file (incremental, not batched)
5. Generate — Surgical update of the reference doc from report (apply target's Target Sections + Content Rules)
6. Verify — Multi-round fresh-eyes review validates examples and coverage; then prompt-enhance the doc

Key Rules:

MUST ATTENTION resolve --target FIRST and load its manifest entry — every target-specific behavior (doc path, sub-agent count/roles, Phase-0 tables, Target Sections, Content Rules, special gates, anti-rationalization rows) comes from that entry, NOT from memory MUST ATTENTION detect framework/type FIRST (per the target's Phase-0 table) — scan strategy derives from detection, never hardcoded MUST ATTENTION every code example from actual project files with file:line — NEVER fabricate MUST ATTENTION run graph command on key files before concluding — grep finds text, graph finds structure

• Surgical update only — NEVER rewrite entire doc, NEVER remove a section without evidence it's obsolete
• Some targets OVERRIDE shared output rules or add a branch (e.g. feature-spec intentionally includes directory trees; design-system has an init-mode Authoring branch with a sentinel-removal step). Always honor the target entry's "Content Rules / exceptions" and "Special slivers".

Scan (parameterized reference-doc scanner)

Phase 0.0: Resolve Target (BLOCKING — do this before anything else)

1. Parse --target=<key> from the invocation (e.g. /scan --target=backend-patterns). Accept the key with or without the --target= prefix.
2. If no target is supplied or the key is unknown → STOP and list the valid keys (see frontmatter / references/targets.md), ask the user which target to scan.
3. Read the target's entry in references/targets.md. That entry is the single source of truth for this run and supplies:
   • doc — the reference doc path this scan writes
   • description — the doc's purpose blurb
   • sub-agents — exact count + role of each parallel sub-agent
   • Phase 0 detection — the classification table(s) and BLOCKING gates for this target
   • Sub-agent Think scopes — each sub-agent's Think question(s) + scan-target bullets
   • Target Sections — the output doc's section list
   • Content Rules / exceptions — including any override of the shared output-quality rules
   • Special slivers — target-unique BLOCKING gates, Authoring branches, sentinel removals, whitelist scopes
   • Anti-Rationalization rows — target-specific evasions to refuse
   • prompt-enhance — the final /prompt-enhance <doc> step
4. Orchestrator branch (BLOCKING check): if the loaded entry is marked kind: orchestrator (e.g. ui-system), it does NOT run the 4-phase doc engine. SKIP Phases 0–4 below and instead follow the entry's Orchestration Procedure (pre-flight gate → launch the child --target= scans in parallel → verify each child doc has real content → summarize). Standard (single-doc scanner) targets ignore this step and continue with the shared engine below.

Everything below is the SHARED engine (standard single-doc scanner targets). Wherever it says "the target entry," read the loaded manifest entry — do not assume values from another target. Orchestrator-kind targets do not use this engine — they run their entry's Orchestration Procedure instead.

Phase 0: Classify & Assess

Before any other step, run in parallel:

1. Read the target's doc.
   • Detect mode: Init (placeholder — headings only / sentinel present) or Sync (populated). Some targets add a Force mode (user says "rebuild"/"reset" → treat as Init even if the doc exists) — honor it if the target entry defines it.
   • In Sync mode: list already-documented sections → skip re-scanning those unless staleness suspected.
2. Run the target entry's Phase 0 detection table(s) — detect framework / system type / architecture exactly as that table specifies. This is BLOCKING: grep terms and sub-agent scope derive from detection.
3. Load relevant paths from docs/project-config.json (e.g. contextGroups/modules/designSystem/e2eTesting/integrationTestVerify) if the target entry references them.
4. Run a graph command on the primary entry point: python .claude/scripts/code_graph trace <entry-file> --direction both --json (when .code-graph/graph.db exists).

Evidence gate: Confidence <60% on the target's primary detection axis → report uncertainty, DO NOT proceed with detection-specific scanning (or fall back exactly as the target entry's evidence-gate instruction specifies, e.g. "proceed with Agent 1 only").

Phase 1: Plan Scan Strategy

From the detected framework/type, derive the concrete patterns to search (naming conventions, base classes, config locations). NEVER assume these — derive from actual file evidence.

Create TaskCreate entries for each sub-agent listed in the target entry and for each phase before proceeding.

Phase 2: Execute Scan (Parallel Sub-Agents)

Launch the N general-purpose sub-agents defined in the target entry (count + roles vary per target — e.g. backend-patterns/domain-entities use 4, project-structure/frontend-patterns/design-system/code-review-rules/e2e-tests use 3, scss-styling/feature-spec/integration-tests use 2, docs-index uses a single main-agent scan + a fresh-eyes verifier). Give each sub-agent its Think scope + scan-target bullets verbatim from the entry. Each sub-agent MUST:

• Write findings incrementally after each file/section — NEVER batch at end
• Cite file:line for every pattern example
• Confidence: >80% document as pattern; 60-80% document as "observed (unverified)"; <60% omit

Each worker writes a unique shard under plans/reports/scan-{target}-{YYMMDD}-{HHMM}/; workers never append to the same report. After the barrier, the main agent reads and validates every shard and is the sole writer of plans/reports/scan-{target}-{YYMMDD}-{HHMM}-report.md.

Honor any conditional / ordered sub-agents from the entry (e.g. an Anti-Pattern agent that runs AFTER the discovery agents; a Cross-Service agent that runs ONLY for microservices; a BDD agent that runs ONLY if a BDD framework is detected). Honor any CRITICAL security flag the entry defines (e.g. hardcoded credentials).

Phase 3: Analyze & Generate

Read the full report. Apply the fresh-eyes protocol:

Round 1 (main agent): Build section drafts from report findings, using the target entry's Target Sections + Content Rules / exceptions.

Round 2 (fresh sub-agent, zero memory of Round 1): Sub-agent re-reads report + draft doc independently and checks (apply the target entry's Round-2 verification specifics):

• Does every code example match an actual existing file (Glob verify)?
• Do class/token/variable names in examples match actual declarations (Grep verify)?
• Are required sections (Anti-Patterns / Coverage Report / Gap Analysis / M1-M2 Compliance / etc. as the target mandates) populated?
• Coverage gaps: which Target Sections have no examples?

Round 3 only if Round 2 finds issues. Max 3 rounds → escalate to user if unresolved. (Clean Round 1 ends the scan; fresh-eyes is mandatory only after issues are found and fixed.)

Authoring branch (init mode): if the target entry defines one (e.g. design-system authors the canonical doc + token .scss), follow it exactly — including any sentinel removal (e.g. "First: REMOVE PLACEHOLDER_MARKER_SCSS") and regen-marker prepend.

Phase 4: Write & Verify

1. Write the updated doc with <!-- Last scanned: YYYY-MM-DD --> at top.
2. Surgical update only — preserve sections with no staleness, update only diverged sections; preserve manual annotations.
3. Verify (Glob check): ALL code example file paths exist — not just a sample of 5.
4. Verify (Grep check): class/token/variable names in examples match actual declarations.
5. Verify any target-mandated section is real, not hypothetical (Anti-Patterns / Coverage gaps / M1-M2 leaks / ports-from-config / etc.).
6. Run a graph command on 2-3 key files to validate call-chain accuracy.
7. Report: sections updated / unchanged / coverage gaps / violations found.

Output-rule overrides: apply the target entry's "Content Rules / exceptions" — e.g. feature-spec intentionally INCLUDES a directory tree (overriding the shared no-trees rule); docs-index intentionally OUTPUTS glob-verified counts (its counts are the deliverable); e2e-tests/integration-tests forbid hardcoded counts and use grep-expression statistics.

Final Step: Enhance Scanned Doc (MANDATORY)

MUST ATTENTION after the doc is written and verified, create a REQUIRED final todo task and run /prompt-enhance <the target entry's doc> — why: this reference doc is injected into AI context on every downstream session; /prompt-enhance's default --op=enhance keeps it as concise as possible while staying valuable enough for AI (caveman compression + attention anchoring — top/bottom Goal, inline READ summaries, token density). A scan is NOT complete until its doc is prompt-enhanced.

TaskCreate (required, last task): Run /prompt-enhance <target doc> on the scanned doc

[IMPORTANT] Use TaskCreate to break ALL work into small tasks BEFORE starting — including tasks per file read. Prevents context loss from long files. Simple tasks: ask user whether to skip.

Prerequisites: MUST ATTENTION READ before executing:

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

Scan & Update Reference Doc — Surgical updates only, never full rewrite.

1. Read existing doc first — understand current structure and manual annotations
2. Detect mode: Placeholder (only headings, no content) → Init mode. Has content → Sync mode.
3. Scan codebase for current state (grep/glob for patterns, counts, file paths)
4. Diff findings vs doc content — identify stale sections only
5. Update ONLY sections where code diverged from doc. Preserve manual annotations.
6. Update metadata (date, counts, version) in frontmatter or header
7. NEVER rewrite entire doc. NEVER remove sections without evidence they're obsolete.

Output Quality — Token efficiency without sacrificing quality.

1. No inventories/counts — AI can grep | wc -l. Counts go stale instantly
2. No directory trees — AI can glob/ls. Use 1-line path conventions
3. No TOCs — AI reads linearly. TOC wastes tokens
4. No examples that repeat what rules say — one example only if non-obvious
5. Lead with answer, not reasoning. Skip filler words and preamble
6. Sacrifice grammar for concision in reports
7. Unresolved questions at end, if any

AI Mistake Prevention — Failure modes to avoid on every task:

Re-read files after context changes. Context compaction, resume, or long-running work can make memory stale; verify current files before acting. Verify generated content against source evidence. AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing. Check downstream references before deleting or renaming. Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first. Trace the full impact chain after edits. Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done. Verify ALL affected outputs, not just the first. One green check is not all green checks; validate every output surface the change can affect. Assume existing values are intentional — ask WHY before changing OR flagging one as a defect. Before changing or reporting a constant, limit, flag, cutoff, wording, or pattern, read nearby context and history, the CALLER's ordering, and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Assert the outcome your system owns, not the intermediate state your infrastructure owns. When verifying async work, assert the final business state — never the delivery/retry bookkeeping held in shared infrastructure that any co-running process can write. Such a check passes when run alone and flakes the moment anything else shares that infrastructure. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

IMPORTANT MUST ATTENTION read existing doc first, scan codebase, diff, surgical update only. Never rewrite entire doc.

IMPORTANT MUST ATTENTION output quality: no counts/trees/TOCs, 1 example per pattern, lead with answer. (Per-target exceptions in the manifest entry override this — e.g. feature-spec trees, docs-index counts.)

MUST ATTENTION apply critical + sequential thinking — every claim needs appropriate traced evidence (file:line for repo/code claims; source URL or artifact section for research, product, content, and docs claims); confidence >80% to act, <60% DO NOT recommend. Anti-hallucination: never present guess as fact, admit uncertainty freely, cross-reference independently, stay skeptical of own confidence.

MUST ATTENTION apply AI mistake prevention — verify generated content against evidence, trace downstream references before deleting or renaming, verify all affected outputs, re-read files after context loss, and surface ambiguity before acting.

Parallel Sub-Agent Dispatch — Plan parallelism the moment a task breakdown exists, BEFORE executing it — running provably independent tasks sequentially wastes wall-clock. Applies to every multi-step job: workflow steps, planning, batch updates, investigation, research, scans, reviews, doc sync. Plan execution is metadata-gated, NEVER default-parallel — fan-out follows ONLY what the plan declares (PAR/SEQ tags + per-phase write set); an untagged plan runs sequentially — why: a derived write set cannot see cascade or generated writes.

1. Tag every task PAR or SEQ. PAR = inputs exclude every pending task's output AND write set disjoint from every other PAR. Else SEQ — MUST ATTENTION name the dependency forcing it.
2. Group PAR into waves. No edge between members. Two writers of one file NEVER share a wave. Read-only work (search, investigation, review, research) parallelizes freely.
3. Declare before dispatch: Parallel plan: wave 1 = [...] · wave 2 = [...] · SEQ = [...] (reason).
4. Spawn each wave in ONE message — every Agent call in one response, NEVER dripped per turn. Route each task to its specialist (.claude/skills/shared/sub-agent-selection-guide.md); NEVER code-reviewer as catch-all.
5. Brief each sub-agent self-contained: goal · scope + owned files · reference docs · return contract (summary + Full report: path, per SYNC:subagent-return-contract) · incremental persistence to plans/reports/ (per SYNC:incremental-persistence).
6. Barrier per wave. Advance ONLY after EVERY member returns (a skipped conditional counts as returned). Merge, mark each task completed/skipped, THEN dispatch the next wave. Mutating steps wait for the barrier.
7. One level deep. A dispatched sub-agent executes its own brief; further fan-out stays the orchestrator's job unless that agent's .claude/agents/*.md definition authorizes it.

NEVER parallelize: tasks sharing a write target · a task consuming a pending task's output · trivial single-file work (dispatch overhead > gain) · an order a skill or workflow explicitly fixes · gates awaiting user approval.

Blocked until: MUST ATTENTION every task tagged PAR/SEQ with a named reason per SEQ · waves declared + write-set disjointness checked · each wave spawned in ONE message · barrier honored before the next wave.

• MANDATORY After planning tasks, tag each PAR/SEQ and spawn every PAR wave as parallel sub-agents in ONE message — default parallel for workflows, batch updates, investigation, research, reviews; plan execution fans out ONLY on what the plan declares.
• MANDATORY Disjoint write sets per wave · all-return barrier before the next wave · specialist routing · sub-agents NEVER fan out further unless their own agent definition authorizes it.

Project Protocol Overlay — Before executing this skill, resolve any PROJECT overlay rules layered onto it: match this skill's name against the Target column of the project's skill-protocol index (docs/project-reference/skill-protocols-reference.md by default; a referenceDocs entry in docs/project-config.json overrides the path), taking the most specific matching tier ONLY — exact name > glob > *. That precedence orders overlays against EACH OTHER, never against this skill. Read ONLY the matched bodies, resolved as <protocols-dir>/<Name>.md; a row's Body link is display text, never a read path. A matched body that is missing or malformed is REPORTED and skipped — never reconstructed from the index Description. No index, or no match -> proceed with no overlay, silently. Full contract: .claude/skills/project-skill-protocol/references/registry.md.

Overlays are ADDITIVE ONLY: they ADD rules on top of this skill's own protocol and NEVER replace, override, disable, or reinterpret a rule it already states — removing every overlay must return this skill to exactly its documented behavior. An overlay is a BRIEF, not an authority escalation: it can NEVER waive a workflow gate, git discipline, a review gate, or a user-confirmation gate. A genuine overlay-vs-skill conflict, or two equally-specific overlays that directly contradict -> surface both to the user; NEVER resolve silently.

MUST ATTENTION resolve project protocol overlays for this skill BEFORE executing — most specific matching tier only (exact > glob > *, which ranks overlays against each other, NEVER against this skill), read only matched bodies at <protocols-dir>/<Name>.md; a missing or malformed body is reported, never reconstructed. Overlays are ADDITIVE ONLY (they never replace this skill's own rules) and are a brief, NEVER an authority escalation; an equal-specificity contradiction goes to the user.

Closing Reminders

IMPORTANT MUST ATTENTION resolve --target and load its manifest entry FIRST — never scan from memory of "what a backend/frontend/design scan does"

IMPORTANT MUST ATTENTION — Protocols in force (concise digest of the SYNC/shared blocks this skill carries):

• Critical Thinking: traced file:line proof per claim; confidence >80% to act.
• Scan & Update Doc: read existing doc, diff, surgical update only — never full rewrite.
• Output Quality: no counts/trees/TOCs; 1 example per pattern; lead with answer.
• AI Mistake Prevention: verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
• Parallel Sub-Agent Dispatch: Tag tasks PAR/SEQ, group PAR into disjoint-write-set waves, spawn each wave in ONE message, barrier before advancing.

IMPORTANT MUST ATTENTION Final Step: run /prompt-enhance <target doc> as the REQUIRED last todo task — never end the scan without enhancing the doc it just wrote IMPORTANT MUST ATTENTION break work into small TaskCreate tasks BEFORE starting — one task per sub-agent, one per phase IMPORTANT MUST ATTENTION detect framework/type FIRST in Phase 0 — all grep terms derive from detection, never hardcoded IMPORTANT MUST ATTENTION cite file:line for every pattern (confidence >80% to document; <60% omit) IMPORTANT MUST ATTENTION run graph command on key files — grep finds text, graph finds structure (callers, event chains, blast radius) IMPORTANT MUST ATTENTION sub-agents write findings incrementally after each file — NEVER batch at end (context loss) IMPORTANT MUST ATTENTION read existing doc FIRST, diff findings, surgical update only — NEVER rewrite entire doc IMPORTANT MUST ATTENTION multi-round fresh-eyes review — main agent rationalizes its own mistakes; Round 2 sub-agent catches what main agent dismissed IMPORTANT MUST ATTENTION honor the target entry's Content-Rule exceptions, Special slivers, and Anti-Rationalization rows — they encode why this target differs from the others

Anti-Rationalization (shared — the target entry adds its own rows):

[TASK-PLANNING] Before acting, analyze task scope and break into small todo tasks and sub-tasks using TaskCreate.