trend-report

Trend Report

Generate a strategic TIPS trend report from agreed trend-scout candidates. Organizes the report around investment themes — each investment theme tells a CxO-level story backed by T→I→P value chain evidence. Dispatches 4 parallel agents to enrich trends with web-sourced quantitative evidence, then assembles a theme-first strategic report with executive summary and claims registry.

Purpose

Transform agreed trend-scout candidates into a strategic, evidence-backed report:

1. Load value-modeler investment themes and validate prerequisites
2. User selects a report-level narrative arc (from cogni-narrative's 7 arcs) to frame the overall story
3. Enrich each trend with quantitative evidence from web research
4. Assemble investment theme narratives using theme-thesis arc (Why Change → Why Now → Why You → Why Pay) with embedded evidence
5. Generate arc-framed executive summary, bridge paragraphs between themes, and synthesis closing section
6. Generate inline citations for every quantitative claim
7. Produce a claims registry compatible with cogni-claims:claims
8. Optionally verify claims via cogni-claims:claims
9. If claims have deviations, revise the report with corrections and removals (Phase 5)
10. Recommend downstream polish, visualization, and export options via /trends-resume

Language Support

Full German and English support. This skill follows the shared language resolution pattern — see $CLAUDE_PLUGIN_ROOT/references/language-resolution.md.

Two language concepts:

1. Interaction language — how the skill communicates with the user (prompts, status, questions). Determined by workspace .workspace-config.json language setting. All Phase 0 prompts, status messages, and error messages use this language.
2. Output language — what language the report is written in. Default priority: (1) trend-scout project_language, (2) workspace language, (3) en. Always ask the user to confirm or override at the start of Phase 0.

Report prose, section headers, and TIPS labels all adapt to the chosen output language. Web searches run bilingually for maximum coverage. German text uses proper umlauts (never ASCII transliterations).

Prerequisites

• trend-scout completed with execution.workflow_state == "agreed" and 60 candidates
• value-modeler completed with tips-value-model.json containing investment themes
• Web access enabled for evidence enrichment
• Optional: cogni-narrative plugin for theme-thesis arc guidance (graceful fallback if absent — investment themes use flat structure)
• Optional: cogni-copywriting plugin for executive polish (graceful fallback if absent)
• Optional: cogni-claims plugin for claim verification

Context Independence

This skill reads ALL required state from project files — it does not depend on prior conversation context. The trends-resume dashboard, earlier questions, and any preceding chat are not inputs to the report pipeline. This means context compaction is safe and recommended before starting.

Before executing Phase 0, run /compact to free working memory. Phase 2 delegates investment theme section writing to parallel agents (reducing orchestrator context from ~69% to ~25-35%), but the orchestrator still reads the value model and claims files for assembly sections. Compacting early ensures headroom.

If /compact is unavailable (e.g., non-interactive mode), proceed without it — Phase 2's agent-based architecture is designed to stay within context limits.

Path Variables

Variable	Purpose	Example
CLAUDE_PLUGIN_ROOT	Plugin installation (skills, references)	~/.claude/plugins/marketplaces/cogni-trends
PROJECT_AGENTS_OPS_ROOT	Workspace root where projects live (optional, set by cogni-workspace)	User's workspace directory

CLAUDE_PLUGIN_ROOT is injected automatically from settings.local.json. PROJECT_AGENTS_OPS_ROOT is set by cogni-workspace — if not present, scripts fall back to $PWD.

Shell Usage

This skill is a pure orchestrator. All file I/O uses Read/Write tools; web research is delegated to agents. The only shell commands needed are:

• cat file1 file2 ... > output — concatenation of log files into the final report
• rm -f pattern — cleanup of stale output files on re-run
• [ -f file ] — existence checks before concatenation

Avoid jq, sed, awk, or grep for data processing — parsing JSON through the LLM keeps the workflow self-contained and avoids dependency issues.

References Index

Read references only when needed for the specific phase:

Reference	Read when...
$CLAUDE_PLUGIN_ROOT/references/language-resolution.md	Language detection and resolution pattern
$CLAUDE_PLUGIN_ROOT/references/data-model.md	Understanding entity schemas and project structure
references/report-arc-frames.md	Arc-specific framing templates for exec summary, bridges, synthesis (Phase 0.4b + Phase 2)
references/phase-2-strategic-themes.md	Assembling the investment theme report (Phase 2)
references/report-structure.md	Dimension section templates (written by Phase 1 agents)
references/evidence-enrichment.md	Configuring agent web search strategy (Phase 1)
references/claims-format.md	Extracting/merging claims (Phase 1-2)
references/i18n/labels-en.md	English report headings and labels
references/i18n/labels-de.md	German report headings and labels
references/phase-3-claim-verification.md	Running claim verification (Phase 3)
references/phase-3.5-executive-polish.md	Polishing report prose (Phase 3.5)
references/phase-5-revision.md	Post-verification report revision (Phase 5)

Workflow Overview

Track progress through these phases as you go:

Phase 0 → Phase 0.5 → Phase 1 → Phase 2 → Phase 2.5 → Phase 3 → Phase 3.5 → Phase 4 → Phase 5
   │          │           │          │          │           │          │            │          │
   │          │           │          │          │           │          │            │          └─ Post-verification revision
   │          │           │          │          │           │          │            └─ Update metadata, recommend /trends-resume
   │          │           │          │          │           │          └─ Optional executive polish
   │          │           │          │          │           └─ Optional claims verification
   │          │           │          │          └─ Structural review (cross-theme quality gate)
   │          │           │          └─ Theme narratives + exec summary + bridges + synthesis
   │          │           └─ 4 parallel agents: enrich trends, write sections, extract claims
   │          └─ Optional deep research for 3-5 high-value ACT-horizon trends
   └─ Project discovery, arc selection, load inputs, validate gate

---

Phase 0: Project Discovery + Input Loading

Step 0.0: Detect Interaction Language

Read $CLAUDE_PLUGIN_ROOT/references/language-resolution.md. Detect workspace language from .workspace-config.json (via ${PROJECT_AGENTS_OPS_ROOT}/.workspace-config.json or CWD). Set INTERACTION_LANGUAGE — use this for all user-facing messages, prompts, and status updates from this point on.

Step 0.1: Project Discovery

Trend-scout projects use trend-scout-output.json (not sprint-log.json), so the shared project-picker.md pattern does not apply.

1. If --project-path was provided as argument, use it directly
2. Otherwise, run discover-projects.sh --json to enumerate all projects
3. For each project, check if {path}/.metadata/trend-scout-output.json exists
4. Read the file and check execution.workflow_state == "agreed" and tips_candidates.total >= 60
5. Collect eligible projects:
   • 0 eligible: ERROR — "No agreed trend-scout projects found. Run trend-scout first."
   • 1 eligible: Auto-select
   • 2+ eligible: Present via AskUserQuestion

Step 0.2: Load Input Data

REQUIRED (validate only — do NOT hold candidates or signals in context):
  {PROJECT_PATH}/.metadata/trend-scout-output.json
    → Extract: config.industry, config.research_topic
    → Extract: config.market_region (default: "dach" if absent — older projects pre-regionalization)
    → Extract: project_language (top-level, NOT config.language)
    → Validate: tips_candidates.total >= 60, execution.workflow_state == "agreed"
    → Do NOT extract tips_candidates.items — agents read these themselves

REQUIRED (value model — keep in context for Phase 2):
  {PROJECT_PATH}/tips-value-model.json
    → Check: investment_themes[] array exists and has ≥1 entry
    → Extract: investment_themes[], value_chains[], solution_templates[]

NOTE: Raw web signals (web-research-raw.json) are NOT loaded by the orchestrator.
Phase 1 agents read and filter signals themselves — see trend-report-writer agent.

Display to the user: "{PHASE_0_INVESTMENT_THEMES_FOUND}" (from i18n labels)

Step 0.2b: Extract Phase 2 Value-Model Subset

The full tips-value-model.json contains scoring matrices, blueprints, and reanchor logs that Phase 2 does not need. To reduce context pressure, extract only the fields Phase 2 uses and write a pruned subset.

Read tips-value-model.json (already loaded in Step 0.2). Write {PROJECT_PATH}/.logs/phase2-value-model.json containing ONLY these top-level keys:

{
  "investment_themes": [],
  "value_chains": [],
  "orphan_candidates": [],
  "coverage": {},
  "mece_validation": {},
  "solution_templates": [
    { "st_id": "...", "name": "...", "category": "...", "enabler_type": "...", "investment_theme_ref": "...", "portfolio_grounding": [...] }
  ]
}

• Copy investment_themes, value_chains, orphan_candidates, coverage, mece_validation in full
• For each solution_templates[] entry, keep ONLY: st_id, name, category, enabler_type, investment_theme_ref, portfolio_grounding — omit solution_blueprint, description, and all other fields. portfolio_grounding is needed for Phase 2 portfolio close (product names and links)
• Omit all other top-level keys (reanchor_log, solution_process_improvements, metrics, collaterals, portfolio_gaps, etc.)

Step 0.3: Validate Entry Gate

Check	Condition	On Failure
Output exists	.metadata/trend-scout-output.json	HALT: Run trend-scout first
Workflow state	== "agreed"	HALT: Complete trend-scout selection
Candidate count	>= 60	HALT: Expected 60 agreed candidates
Value model exists	tips-value-model.json with investment_themes[]	HALT: Run value-modeler first
Config complete	industry, subsector, language present	HALT: Incomplete config

Step 0.4: Ask User for Deliverable Language

The project_language from trend-scout-output.json is the default (falling back to workspace language if not set). Always confirm with the user. Present the question in the INTERACTION_LANGUAGE:

If INTERACTION_LANGUAGE == "de":

AskUserQuestion:
  question: "In welcher Sprache soll der Report erstellt werden? trend-scout hat '{project_language}' verwendet."
  header: "Report-Sprache"
  options:
    - label: "Deutsch (DE) ← Standard"
    - label: "English (EN)"

If INTERACTION_LANGUAGE == "en":

AskUserQuestion:
  question: "Report language? trend-scout used '{project_language}'. Keep or change?"
  header: "Report language"
  options:
    - label: "English (EN) ← Default"
    - label: "Deutsch (DE)"

The option matching the current default gets the arrow marker. Set LANGUAGE to the user's choice. Update project_language in trend-scout-output.json if changed.

Step 0.4b: Select Report-Level Narrative Arc

Read references/report-arc-frames.md for the full arc frame definitions.

The report-level arc determines how investment themes connect into one cohesive narrative — the executive summary voice, bridge paragraphs between themes, and the synthesis closing section. It does NOT change how individual theme sections are written internally (those always use the theme-thesis arc).

Present the 7 available arcs via AskUserQuestion. The recommended default is corporate-visions (the proven B2B persuasion frame). Auto-detect a different recommendation if the topic strongly signals another arc (e.g., heavily regulatory topics → industry-transformation).

If INTERACTION_LANGUAGE == "de":

AskUserQuestion:
  question: "{PHASE_0_ARC_QUESTION}"
  header: "{PHASE_0_ARC_HEADER}"
  options:
    - label: "{ARC_CORPORATE_VISIONS}"
      description: "{ARC_CORPORATE_VISIONS_DESC}"
    - label: "{ARC_TECHNOLOGY_FUTURES}"
      description: "{ARC_TECHNOLOGY_FUTURES_DESC}"
    - label: "{ARC_INDUSTRY_TRANSFORMATION}"
      description: "{ARC_INDUSTRY_TRANSFORMATION_DESC}"
    - label: "{ARC_STRATEGIC_FORESIGHT}"
      description: "{ARC_STRATEGIC_FORESIGHT_DESC}"

If INTERACTION_LANGUAGE == "en":

AskUserQuestion:
  question: "{PHASE_0_ARC_QUESTION}"
  header: "{PHASE_0_ARC_HEADER}"
  options:
    - label: "{ARC_CORPORATE_VISIONS}"
      description: "{ARC_CORPORATE_VISIONS_DESC}"
    - label: "{ARC_TECHNOLOGY_FUTURES}"
      description: "{ARC_TECHNOLOGY_FUTURES_DESC}"
    - label: "{ARC_INDUSTRY_TRANSFORMATION}"
      description: "{ARC_INDUSTRY_TRANSFORMATION_DESC}"
    - label: "{ARC_STRATEGIC_FORESIGHT}"
      description: "{ARC_STRATEGIC_FORESIGHT_DESC}"

AskUserQuestion limit: The picker supports max 4 options. Present the 4 most relevant arcs based on the topic. The recommended arc is always first (with arrow marker). If the user selects "Other", show the remaining 3 arcs (competitive-intelligence, trend-panorama, theme-thesis) in a follow-up question.

Set REPORT_ARC_ID to the user's choice (e.g., corporate-visions, technology-futures). This variable is passed to Phase 2 for arc-aware assembly.

Step 0.4c: Propose Report Title

The research question ({TOPIC}) becomes the subtitle. Generate a punchy title (max 8 words) from {TOPIC}, REPORT_ARC_ID, INDUSTRY, and investment theme names (already loaded in Step 0.2). The title should:

• Be CxO-level — sharp, memorable, forward-looking
• Reflect the arc's rhetorical stance (e.g., corporate-visions → urgency/challenge; technology-futures → capability/convergence; industry-transformation → structural shift)
• NOT repeat the research question or be a generic label like "Trend Report"

The subtitle is {TOPIC}, optionally shortened for readability (drop redundant geographic/industry qualifiers if obvious from context).

Present via AskUserQuestion:

If INTERACTION_LANGUAGE == "de":

AskUserQuestion:
  question: "{PHASE_0_TITLE_QUESTION}\n\n**Titel:** {proposed_title}\n**Untertitel:** {proposed_subtitle}"
  header: "{PHASE_0_TITLE_HEADER}"
  options:
    - label: "{PHASE_0_TITLE_ACCEPT}"
    - label: "{PHASE_0_TITLE_EDIT}"

If INTERACTION_LANGUAGE == "en":

AskUserQuestion:
  question: "{PHASE_0_TITLE_QUESTION}\n\n**Title:** {proposed_title}\n**Subtitle:** {proposed_subtitle}"
  header: "{PHASE_0_TITLE_HEADER}"
  options:
    - label: "{PHASE_0_TITLE_ACCEPT}"
    - label: "{PHASE_0_TITLE_EDIT}"

If the user selects Edit, ask them to provide their preferred title and/or subtitle as free text. Store the final values as {TITLE} and {SUBTITLE}.

Step 0.5: Load i18n Labels

Read the labels file matching the chosen language:

• English: references/i18n/labels-en.md
• German: references/i18n/labels-de.md

Step 0.6: Clean Up Stale Output Files

On re-runs, remove stale files to prevent mixing old and new content:

rm -f "{PROJECT_PATH}/.logs/report-header.md" \
      "{PROJECT_PATH}/.logs/report-section-"*.md \
      "{PROJECT_PATH}/.logs/report-investment-theme-"*.md \
      "{PROJECT_PATH}/.logs/report-bridge-"*.md \
      "{PROJECT_PATH}/.logs/report-synthesis.md" \
      "{PROJECT_PATH}/.logs/enriched-trends-"*.json \
      "{PROJECT_PATH}/.logs/claims-"*.json \
      "{PROJECT_PATH}/.logs/report-claims-registry.md" \
      "{PROJECT_PATH}/tips-trend-report.md" \
      "{PROJECT_PATH}/.logs/phase2-value-model.json" \
      "{PROJECT_PATH}/tips-trend-report-claims.json" \
      "{PROJECT_PATH}/tips-insight-summary.md"

---

Phase 0.5: Deep Research Selection (Optional)

When to run: Offer this phase when the user wants deeper evidence for high-value trends. Skip if the user wants a fast report or explicitly declines.

Purpose: Select 3-5 high-value ACT-horizon trends for recursive deep research before standard evidence enrichment. Deep-researched trends get richer evidence (quantitative data, forcing functions, ROI figures) that makes the Why Change / Why Pay investment theme arguments substantially more credible.

Step 0.5.1: Select Trends for Deep Research

Ask the user via AskUserQuestion:

EN: "I can perform deep research on 3-5 high-value trends before writing the report. This adds ~5-10 minutes but produces richer evidence with quantitative data. Would you like to:
a) Deep research top ACT-horizon trends (recommended for executive audiences)
b) Skip deep research and proceed with standard evidence enrichment
c) Select specific trends for deep research"

DE: "Ich kann eine Tiefenrecherche für 3-5 hochwertige Trends durchführen, bevor der Bericht geschrieben wird. Das dauert ~5-10 Minuten länger, liefert aber reichere Evidenz mit quantitativen Daten. Möchten Sie:
a) Tiefenrecherche der wichtigsten ACT-Horizont-Trends (empfohlen für Führungskräfte-Publikum)
b) Tiefenrecherche überspringen und mit Standard-Evidenzanreicherung fortfahren
c) Spezifische Trends für Tiefenrecherche auswählen"

Step 0.5.2: Auto-Select or User-Select

If user chose (a), auto-select using these criteria (in priority order):

1. ACT-horizon trends with signal_intensity >= 4 and confidence_tier == "high" — sorted by composite score descending
2. If fewer than 3 qualify, include ACT-horizon trends with confidence_tier == "medium" and highest scores
3. Cap at 5 trends maximum

If user chose (c), present the ACT-horizon trend list and let them pick.

Step 0.5.3: Dispatch Deep Researchers (Parallel)

Dispatch one trend-deep-researcher agent per selected trend, all in parallel:

Task:
  subagent_type: "cogni-trends:trend-deep-researcher"
  description: "Deep research: {TREND_NAME}"
  prompt: |
    Perform deep research on this trend candidate.

    PROJECT_PATH: {{PROJECT_PATH}}
    TREND_NAME: {{TREND_NAME}}
    TREND_KEYWORDS: {{KEYWORDS}}
    DIMENSION: {{DIMENSION}}
    HORIZON: act
    SUBSECTOR_EN: {{SUBSECTOR_EN}}
    SUBSECTOR_DE: {{SUBSECTOR_DE}}
    RESEARCH_HINT: {{RESEARCH_HINT}}
    MARKET_REGION: {{MARKET_REGION}}

Process results: Each agent writes a .logs/deep-research-{slug}.json artifact. Log success/failure counts. These artifacts are consumed by Phase 1 trend-report-writer agents — trends with deep research artifacts skip their own WebSearch and use the richer findings directly.

---

Phase 1: Evidence Enrichment + Section Generation (PARALLEL)

Read references/evidence-enrichment.md for web search strategy. Read references/claims-format.md for claims extraction schema.

Deep research integration: Before each trend-report-writer agent runs its evidence enrichment, it checks for {PROJECT_PATH}/.logs/deep-research-{trend-slug}.json. If a deep research artifact exists for a trend in its dimension, the writer uses the artifact's synthesis and sources instead of running its own WebSearch for that trend. This is a fourth evidence status alongside signal_sufficient, signal_partial, and signal_none: deep_research_available — richest evidence tier, no additional search needed.

Step 1.1: Dispatch 4 Agents

Dispatch all 4 agents in a single message (parallel tool calls) so they run concurrently:

Per agent:
  subagent_type: "cogni-trends:trend-report-writer"
  model: sonnet
  prompt: |
    Dimension: {DIMENSION}
    TIPS Role: {TIPS_ROLE}
    Project Path: {PROJECT_PATH}
    Language: {LANGUAGE}
    Market Region: {MARKET_REGION}
    Industry EN/DE: {INDUSTRY_EN} / {INDUSTRY_DE}
    Subsector EN/DE: {SUBSECTOR_EN} / {SUBSECTOR_DE}
    Topic: {TOPIC}
    Labels: {relevant i18n labels}

Agents self-load candidates and raw signals from disk using PROJECT_PATH — no need to pass data in the prompt. This keeps the orchestrator context lean for Phase 2.

Dimensions: externe-effekte (T), digitale-wertetreiber (I), neue-horizonte (P), digitales-fundament (S).

Each agent writes:

• {PROJECT_PATH}/.logs/report-section-{dimension}.md — narrative section (dimension-level prose)
• {PROJECT_PATH}/.logs/claims-{dimension}.json — extracted claims
• {PROJECT_PATH}/.logs/enriched-trends-{dimension}.json — per-trend evidence blocks keyed by candidate_ref; actions_md uses semicolon-separated keywords (used in investment theme assembly)

Step 1.2: Collect Agent Results

Each agent returns compact JSON with ok, dimension, trends_covered, claims_extracted, signal reuse metrics, and the three output file paths (section_file, claims_file, enriched_file).

If an agent returns ok: false: retry once. If retry also fails: HALT with the dimension name. All 4 must succeed before Phase 2.

Step 1.3: Validate Agent Output Files

After all 4 agents complete, verify that all 12 expected files exist:

For each dimension in [externe-effekte, digitale-wertetreiber, neue-horizonte, digitales-fundament]:
  ✓ {PROJECT_PATH}/.logs/report-section-{dimension}.md    — narrative section (intermediate artifact)
  ✓ {PROJECT_PATH}/.logs/claims-{dimension}.json           — extracted claims
  ✓ {PROJECT_PATH}/.logs/enriched-trends-{dimension}.json  — per-trend evidence blocks (required for Phase 2)

If any report-section-{dimension}.md file is missing, log a WARNING. Phase 2 can proceed (it uses enriched-trends).

---

Phase 2: Report Assembly — THEME-FIRST (NOT BY DIMENSION)

CRITICAL: You MUST read references/phase-2-strategic-themes.md before starting Phase 2. The report is organized by investment themes from tips-value-model.json, NOT by TIPS dimension. Do NOT simply concatenate the dimension section files from Phase 1 — those are intermediate artifacts, not the final report structure.

EXECUTION — Agent-assisted investment theme writing + orchestrator assembly. Phase 2 delegates the context-heavy investment theme section writing to parallel trend-report-investment-theme-writer agents (one per investment theme), then the orchestrator writes the remaining lightweight sections (executive summary, claims registry) and concatenates the final report.

• Investment theme sections → agents. Each investment-theme-writer agent self-loads enriched-trends and claims from disk, filtered to its own candidate_refs. The orchestrator passes only small scalars (investment theme definition, value chains, labels). This keeps the orchestrator's context lean.
• No Python scripts. Do not write scripts to generate report sections. The agents and orchestrator write strategic prose directly.
• No intermediate analysis steps. Do not generate lookup documentation or enriched statistics. Agents go straight from reading data to writing prose.

Summary of steps (details in the reference):

1. Read value model — Read .logs/phase2-value-model.json for investment themes, value chains, solution templates, orphan candidates, coverage data
2. Dispatch investment theme agents — For each investment theme, dispatch a cogni-trends:trend-report-investment-theme-writer agent with MARKET_REGION: {MARKET_REGION} and REPORT_ARC_ID: {REPORT_ARC_ID} in the prompt. All agents in a single message (parallel). Each agent self-loads evidence from disk, writes report-investment-theme-{investment_theme_id}.md, and returns compact JSON with word count, citation count, quality gate status, and top claims.
3. Collect agent results — Validate all agents returned ok: true and quality gates passed. Retry once on failure.
4. Write executive summary — Read ALL report-investment-theme-{investment_theme_id}.md files. Use REPORT_ARC_ID to select the arc-specific opener and closer patterns from report-arc-frames.md. Write report-header.md.
5. Write bridge paragraphs — For each consecutive theme pair, generate a 2-4 sentence bridge using the arc's bridge pattern. Write report-bridge-{N}-{N+1}.md files.
6. Write synthesis section — Generate a 300-500 word closing section using the arc's synthesis frame. Aggregates evidence across all themes. Write report-synthesis.md.
7. Write claims registry — Read 4 claims-{dimension}.json files once, map claims to investment themes via value model, write report-claims-registry.md
8. Assemble — Concatenate: header + (theme1 + bridge-1-2 + theme2 + bridge-2-3 + ... + themeN) + synthesis + claims → tips-trend-report.md
9. Merge claims → tips-trend-report-claims.json

Resume logic: Before dispatching an agent for an investment theme, check if report-investment-theme-{investment_theme_id}.md already exists and is >1000 bytes. If so, skip that agent — display "{PHASE_2_INVESTMENT_THEME_AGENT_SKIP_RESUME}" and continue. This means re-runs only dispatch for missing investment themes.

---

Phase 2.5: Structural Review (Optional but Recommended)

Purpose: Cross-theme quality check of the assembled report before claims verification. Individual agents have their own quality gates, but this phase catches issues that span themes — duplicate evidence, inconsistent forcing functions, themes with zero quantitative data, missing Handeln/Nichthandeln contrasts.

Step 2.5.1: Dispatch Reviewer Agent

Task:
  subagent_type: "cogni-trends:trend-report-reviewer"
  description: "Structural review of trend report"
  prompt: |
    Review the assembled trend report for structural quality.

    PROJECT_PATH: {{PROJECT_PATH}}
    REPORT_PATH: {{PROJECT_PATH}}/tips-trend-report.md
    REVIEW_ITERATION: 1
    OUTPUT_LANGUAGE: {{PROJECT_LANGUAGE}}

Step 2.5.2: Process Verdict

If verdict is "accept" (score >= 0.80): proceed to Phase 3.

If verdict is "revise":

1. Read revision_priorities from the verdict
2. Apply targeted fixes to tips-trend-report.md:
   • Add missing quantitative evidence (may require 1-2 WebSearch calls for specific data points)
   • Add Nichthandeln contrasts where flagged
   • Resolve forcing function inconsistencies
   • Fix duplicate evidence citations
3. Re-run reviewer with REVIEW_ITERATION: 2
4. Accept regardless of second verdict (max 2 iterations) — log remaining issues as warnings

Skip condition: If the user explicitly asks for a fast report or says "skip review", bypass this phase.

---

Phase 3: Claim Verification (Optional)

Read references/phase-3-claim-verification.md for the full workflow. Asks the user whether to verify extracted claims via cogni-claims:claims. If the plugin is not installed, skip with a warning.

---

Phase 3.5: Executive Polish (Optional)

Read references/phase-3.5-executive-polish.md for the full workflow. Polishes report prose via cogni-copywriting:copywriter with SCOPE=tone. Validates citations and structure are preserved; reverts on failure.

---

Phase 4: Finalization

Step 4.1: Update Metadata

Update {PROJECT_PATH}/tips-project.json with current timestamp:

{ "updated": "ISO-8601" }

Add to {PROJECT_PATH}/.metadata/trend-scout-output.json:

{
  "trend_report_complete": true,
  "trend_report_path": "tips-trend-report.md",
  "trend_report_claims_path": "tips-trend-report-claims.json",
  "trend_report_mode": "strategic-themes",
  "trend_report_investment_theme_count": N,
  "trend_report_generated_at": "ISO-8601"
}

Step 4.2: Display Summary

Trend Report Complete (Investment Themes)
─────────────────────────────────────────
Report:       {PROJECT_PATH}/tips-trend-report.md
Themes:       {N} investment themes ({REPORT_ARC_ID} arc)
Claims:       {PROJECT_PATH}/tips-trend-report-claims.json
Trends:       60 across {N} investment themes
Claims:       {total_claims} quantitative claims extracted
Verification: {verdict or "skipped"}

Run /trends-resume to see your full options — polish, visualize, present, export, or accumulate.

---

Phase 5: Post-Verification Revision (Conditional)

Read references/phase-5-revision.md for the full workflow. Triggers when claims verification (Phase 3) or a later /claims session produced resolved deviations that require report corrections or claim removals. If all claims verified clean or the user skipped verification, this phase is skipped.

Dispatches the trend-report-revisor agent to:

• Remove unverifiable claims from the report body and claims registry table (no strikethrough — clean removal)
• Correct inaccurate claims with verified replacement text
• Find replacement evidence via WebSearch when removed claims leave gaps in the argument
• Renumber the claims registry table sequentially
• Output a versioned revision (tips-trend-report-v2.md) that becomes the new canonical report

This phase also handles the deferred flow: when the user verified and resolved claims in a separate session via /claims and returns via /trends-resume, the resume skill detects the revision phase and routes here.

---

Error Handling

Scenario	Action
trend-scout-output.json missing	HALT: Run trend-scout first
workflow_state != "agreed"	HALT: Complete candidate selection
tips_candidates.total < 60	HALT: Expected 60 candidates
tips-value-model.json missing or no investment themes	HALT: Run value-modeler first
tips-value-model.json has investment themes but no value chains	HALT: value-modeler Phase 1 incomplete
No raw signals file (both sources)	WARNING: proceed without signals (~120 searches)
Phase 1 agent returns ok: false	Retry once, then HALT with dimension name
All 4 Phase 1 agents fail	HALT: Check web access is enabled
enriched-trends JSON missing	HALT: Phase 1 agent failed to produce enriched output
Investment theme agent returns ok: false	Retry once, then HALT with investment theme name
Investment theme agent quality gate fails	WARNING: continue (section written but may be thin)
Investment theme references unknown candidate_ref	WARNING: agent skips that candidate in investment theme narrative
cogni-narrative not installed	WARNING: investment-theme-writer uses flat structure (no arc guidance)
cogni-copywriting not installed	WARNING: skip executive polish
cogni-claims not installed	WARNING: skip verification
claims verification returns FAIL	Present failed claims. Do not auto-correct.

Integration

Upstream:

• trend-scout produces trend-scout-output.json (required)
• value-modeler produces tips-value-model.json (required)

Pipeline: trend-scout → value-modeler → trend-report

Optional cross-plugin: cogni-narrative theme-thesis arc (Phase 2 investment theme writer guidance), cogni-claims:claims (Phase 3)

Downstream (via /trends-resume): cogni-copywriting:copywrite (prose polish), cogni-visual:enrich-report (themed HTML), cogni-visual:story-to-slides (presentation), cogni-visual:story-to-web (landing page), cogni-visual:story-to-big-picture (journey map), cogni-visual:story-to-storyboard (print posters), trends-catalog import, trends-dashboard

Debugging

Log files in {PROJECT_PATH}/.logs/:

• report-header.md — frontmatter + exec summary
• report-section-{dimension}.md — dimension sections (4 files, written by agents)
• phase2-value-model.json — pruned value-model subset for Phase 2
• enriched-trends-{dimension}.json — per-trend evidence blocks (4 files, used in investment theme assembly)
• report-investment-theme-{investment_theme_id}.md — investment theme sections (3-7 files, written by investment theme agents)
• claims-{dimension}.json — dimension claims (4 files)
• report-claims-registry.md — claims table

Output files in {PROJECT_PATH}/:

• tips-trend-report.md — assembled final report
• tips-trend-report-claims.json — merged claims registry
• tips-insight-summary.md — legacy artifact (no longer generated; cleaned up on re-runs)

Issue	Check
Phase 1 agent hangs	Verify web access is enabled
Investment theme agent hangs	Check enriched-trends files exist in .logs/
Empty claims	Check if trends have quantitative data in trend-scout output
Wrong language	Verify project_language in trend-scout-output.json
Missing sections	Check .logs/ for partial agent output