# DO NOT EDIT -- overwritten on every update.
#
# Workflow customization surface for bmad-deep-recon.
#
# Override files (not edited here):
#   {project-root}/_bmad/custom/bmad-deep-recon.toml         (team)
#   {project-root}/_bmad/custom/bmad-deep-recon.user.toml    (personal)

[workflow]

# --- Configurable below. Overrides merge per BMad structural rules: ---
#   scalars: override wins
#   arrays (persistent_facts, activation_steps_*, *_sources, doc_standards,
#   external_*): append
#   arrays of tables keyed by `code`: matching key replaces, new keys append

# Steps executed on activation: prepend runs before the skill's own
# activation flow, append runs after it. Each entry is a literal instruction.
activation_steps_prepend = []
activation_steps_append = []

# Standing context for framing the research — decision context only, never
# evidence: the research firewall keeps project material out of findings.
# Entries prefixed `file:` are paths or globs whose contents load as facts;
# all others are literal facts. Empty by default so nothing local leaks into
# research framing unasked.
persistent_facts = []

# Where research runs live and how each run folder is named. Draft, Process,
# and Run all use the same folder shape: brief.md (drafted prompts),
# imports/ (originals, full fidelity), digests/ (extracted claims),
# research.md (the canonical summary/report), .memlog.md.
research_output_path = "{planning_artifacts}/research"
run_folder_pattern = "{research_type}-{topic_slug}-{date}"

# Seed document for a new run.
research_template = "assets/research.template.md"

# --- Effort (Run mode) ------------------------------------------------------
# A preset bundles the three effort knobs; any knob set here individually
# pins that knob over the preset. What the user says in the request beats
# both.
#
#   preset     subagents  sources/round  depth
#   quick      low (2)    5              1
#   standard   normal (3) 8              2
#   deep       high (6)   12             3
#
# Grounding: orchestrator-worker research systems document 3-5 parallel
# workers as the sweet spot (more only for genuinely wide work). Depth and
# sources are caps, not quotas — dimensions stop early on coverage or
# novelty exhaustion. Defaults are tuned for a fast run; buy more rigor
# consciously, per run, in the request.

preset = "standard"

# "" = from preset. Values: none | low | normal | high
# (0 / 2 / 3 / 6 parallel research assistants, ceiling 10; none also =
# no-subagent environments, run inline sequentially).
subagents = ""

# 0 = from preset. Distinct sources actually read per dimension per round.
# Ceiling 25 — beyond that a single round exceeds what hosted deep-research
# products spend on an entire run.
max_sources_per_round = 0

# 0 = from preset. Rounds per dimension: initial pass + lead-following
# follow-ups. Ceiling 5.
max_depth = 0

# Verification level, applied as material lands (never an end-of-run pass).
#   normal  spot-check load-bearing claims only — fast, the default
#   high    cross-check the pack's two-source classes; red-team major
#           conclusions
#   max     cross-check every ledger claim + the red-team pass at full
#           breadth + primary-source-priority ranking
validation = "normal"

# Red-team stance pass — fresh-context skeptics hunting disconfirming
# evidence for major conclusions: "off" (default), "offer" (proposed at the
# plan gate), or "on" (always; headless honors only "on"). high/max
# validation includes it for major conclusions regardless.
red_team = "off"

# Run the acquisition fan-out through the harness's deterministic
# orchestration when it offers one (e.g. workflows): "off", "offer"
# (proposed at the plan gate when available), or "on". Orchestrated runs are
# faster wall-clock but spend more tokens.
use_workflows = "offer"

# Ordered model preference for spawned research assistants — first model the
# harness can provide wins; [] lets the harness/skill choose (lead stays on
# the strongest model; researchers at most one capable tier down; judgment
# work never on the smallest tier; mechanical extraction may use a fast
# tier).
#
# Example: subagent_models = ["<your-mid-tier-model-id>", "<your-fast-model-id>"]
subagent_models = []

# Source policy. Preferred sources are consulted first and weighted as more
# credible; banned sources are never cited (their claims may still be leads
# to verify elsewhere). Entries are domains or plain-text descriptions.
# Draft mode writes both policies into drafted prompts.
preferred_sources = []
banned_sources = []

# What is presented and handed off — never what exists: research.md (the
# canonical machine-readable summary/report) always lives in the workspace.
#   "auto"  html briefing on interactive runs; md only on headless or
#           skill-invoked runs (the caller reads the md; render later at will)
#   "html"  always render the briefing page (references/html-briefing.md)
#   "md"    never render html
#   "both"  render and present both
output_format = "auto"

# Theme for the HTML briefing: empty = the shipped neutral professional
# theme, a `file:` path to a theme/brand spec, or inline directives
# (e.g. "Canvas #122543, accent #B66D46, sans-serif, dark-mode aware").
html_theme = ""

# Default audience shaping for the synthesis — freeform, empty = balanced
# technical/business register. Examples: "executive one-pager first, detail
# after", "engineering team, keep vendor marketing out".
audience = ""

# Registry of extra research surfaces — internal knowledge bases or search
# tools you subscribe to — consulted alongside web research in Run mode; each
# entry names the tool and when to use it. Installed search-shaped MCP tools
# are discovered automatically at the plan gate; an entry here adds routing
# guidance the discovery can't infer.
#
# Examples:
#   external_sources = [
#     "Tavily MCP (tavily_search/tavily_extract): preferred web search + clean page extraction",
#     "Perplexity Sonar MCP (perplexity_ask): cited synthesized answers — chase its citations as the sources",
#     "xAI X Search MCP: live X/Twitter posts and threads, for user-voice and sentiment dimensions",
#     "Gartner MCP (corp:gartner_query): analyst data on enterprise software markets",
#   ]
external_sources = []

# Polish passes applied to research.md at finalize. Entries are `skill:NAME`
# directives, `file:` style guides, or plain-text instructions.
doc_standards = ["skill:bmad-review lenses=structure,prose"]

# Handoffs executed at finalize to route the report beyond local files. Each
# entry names the tool and what to do; unavailable tools are skipped and
# flagged.
#
# Examples:
#   "NotebookLM (notebooklm-mcp): create a notebook from research.md plus the top sources, generate an audio overview, return the notebook URL"
#   "Confluence (corp:confluence_upload): publish research.md to the RESEARCH space, return the page URL"
external_handoffs = []

# Executed after finalize. A string scalar is one instruction; an array is a
# sequence. Empty = the run ends with the finalize summary.
on_complete = ""

# ---------------------------------------------------------------------------
# Research types — subject lenses. Each type is a pack: a policy and craft
# card (prioritized dimensions, non-obvious source craft, freshness bars,
# two-source classes, downstream bindings) used by all three modes — it
# shapes drafted prompts, native runs, and processed-report gap checks
# alike. `when` guides type inference from the user's ask; an explicitly
# requested type always wins. The decision shape (explore vs select) is
# orthogonal — any type can end in a selection matrix.
#
# Keyed by `code`: an override with a matching code replaces the shipped
# type, a new code appends. Empty `pack` disables a type.
#
# Example (add an org-specific type in team/user override TOML):
#   [[workflow.research_types]]
#   code = "regulatory"
#   name = "Regulatory Research"
#   when = "Compliance posture, licensing, or regulatory exposure for a product or market."
#   pack = "file:{project-root}/_bmad/custom/packs/regulatory.md"
# ---------------------------------------------------------------------------

[[workflow.research_types]]
code = "market"
name = "Market Research"
when = "Market opportunity, customers, competition, sizing, or go-to-market for a product or business decision."
pack = "types/market.md"

[[workflow.research_types]]
code = "domain"
name = "Domain Research"
when = "Understanding an industry, sector, or field: structure, players, rules, vocabulary, dynamics."
pack = "types/domain.md"

[[workflow.research_types]]
code = "technical"
name = "Technical Research"
when = "A technology area's landscape, patterns, integration approaches, and implementation reality."
pack = "types/technical.md"

[[workflow.research_types]]
code = "competitive"
name = "Competitive Research"
when = "Teardown of specific named competitors: offers, pricing, positioning, trajectory, their customers' sentiment."
pack = "types/competitive.md"

[[workflow.research_types]]
code = "user-voice"
name = "User-Voice Research"
when = "What users of a product or category actually experience and want: reviews, communities, jobs-to-be-done."
pack = "types/user-voice.md"

[[workflow.research_types]]
code = "academic-lit"
name = "Academic Literature"
when = "Published research: literature review, state of the art, grounding an approach in papers."
pack = "types/academic-lit.md"