Estimate Prep

• TASK (generic shell mode, no profile)

Phase parameter — single-run discipline

The PHASE parameter determines which pipeline executes. A single invocation runs exactly one phase:

Critical invariant: a single run MUST NOT span both phases — the human gate between them is a non-negotiable decision point. If a brief requests both phases in one run, halt with FAILED_INPUTS.

Inputs

Common (both phases)

Required:

• EXECUTION_ROOT — root of the current execution/workspace
• PHASE — SCAFFOLD | BOE (validated enum; invalid = FAILED_INPUTS)
• DECOMPOSITION_PATH — path to the latest decomposition markdown (PROJECT_DECOMP or SOFTWARE_DECOMP)
• SOURCE_DOCUMENTS — path(s) to source documents (RFP, addenda, specs, reference reports)
• CURRENCY — ISO-like code (e.g., USD, CAD)
• PROJECT_CONTEXT — structured block (Location, BaseYear, ProjectType, ProcurementModel; EstimatedValue optional; AdditionalContext optional)

Optional:

• OUTPUT_LABEL — short label for snapshot naming (default AUTO)
• SECONDARY_SOURCES — path(s) to secondary reference documents
• CANONICAL_PRICESOURCES_ROOT — path to canonical pricing library (e.g., {EXECUTION_ROOT}/_PriceSources/) used for schema discovery and/or as ENRICH input
• SCHEMA_MODE — AUTO_FROM_CANONICAL (default) | DEFAULT_COMPAT
• EXPORT_BUNDLE — MANIFEST_ONLY (default) | MANIFEST_AND_PACKAGE

SCAFFOLD-specific

• MODE — BOOTSTRAP (default) | ENRICH
• PRIOR_SNAPSHOT — required if MODE=ENRICH (or use CANONICAL_PRICESOURCES_ROOT)
• HUMAN_PRICING — path(s) to human-provided pricing (quotes, rate tables, historical data, vendor proposals)
• RATE_SCOPE — PRODUCTION_ONLY (default) | PRODUCTION_AND_CONSTRUCTION
• DISCIPLINE_HINTS — override/supplement discipline detection from decomposition

BOE-specific

Required:

• SCAFFOLD_PATH — path to approved SCAFFOLD snapshot (may have been modified by human after Phase SCAFFOLD)

Optional:

• DEPENDENCY_SOURCES — AUTO (default; reads per-deliverable Dependencies.csv) or explicit path(s) to dependency registers
• EVALUATION_CRITERIA — path to or structured block of evaluation criteria with point allocations
• AGGREGATION_HINTS — human-specified aggregation preferences

All resolved defaults and chosen paths MUST be recorded in the snapshot Run_Context.md.

Runtime overrides

Tool usage

• The allowed-tools frontmatter field is intentionally omitted. The skill dispatches several utility scripts under tools/ (enum validation, snapshot scaffolding, INDEX generation) but their invocation is guided operationally, not enforced through the TASK-consumed allowed-tools contract.
• Preferred utility scripts (invoked by the agent during the run when available):
  • tools/validation/validate_enum.py — validates PHASE input.
  • tools/scaffolding/scaffold_tool_root.sh — creates _EstimatePrep/ tool root.
  • tools/scaffolding/create_snapshot_folder.sh — creates immutable snapshot folder.
  • tools/reporting/generate_index_md.sh — generates file inventory portion of INDEX.md.
• Reasoning + file-write operations generate the content (CSVs, scaffold, BOE) based on the canonical schemas in the Schema Annex.

Disallowed behavior:

• No writes outside {EXECUTION_ROOT}/_EstimatePrep/.
• No overwriting prior snapshots.
• No modification of _PriceSources/, _Estimates/, deliverable folders, decomposition outputs, or dependency registers.
• No recursive ingestion of prior _EstimatePrep/ outputs unless explicitly provided as PRIOR_SNAPSHOT / SCAFFOLD_PATH.

Phase SCAFFOLD — method

1. Resolve tool root + create snapshot folder. Validate PHASE enum; bootstrap _EstimatePrep/; create EPREP_SCAFFOLD_{LABEL}_{DATE}_{TIME}/.
2. Load decomposition. Extract Package IDs, Deliverable IDs, scope items, disciplines, owner roles, variant hints.
3. Load source documents. Extract project name, location, owner, timeline, submission requirements, scope boundaries, referenced standards, constraints. Record TBD for items not found.
4. Determine canonical schemas. If SCHEMA_MODE=AUTO_FROM_CANONICAL and CANONICAL_PRICESOURCES_ROOT has a matching file, use its header exactly; otherwise use the Schema Annex (see SCHEMA_ANNEX.md for the 9 schema families).
5. Generate Professional_Staff_Rates.csv (Family 4). Default Confidence=MEDIUM, Basis=PARAMETRIC. ENRICH mode overlays human rates and logs each override.
6. Generate Level_of_Effort.csv (Family 9). Default Basis=PARAMETRIC. No Confidence column — provenance is Basis alone.
7. Generate Assumed_Project_Parameters.csv (Family 8). Uses Source column (not Basis); canonical enum: ASSUMPTION, DESIGN_BASIS, CONFIRMED, DERIVED, PARAMETRIC.
8. Generate additional rate tables (conditional). If RATE_SCOPE=PRODUCTION_AND_CONSTRUCTION, generate all applicable Family 1-7 files matching exact canonical schemas.
9. Generate PriceSources/INDEX.md with the 7 required sections: header block, data quality statement (HIGH/MEDIUM/LOW definitions), file inventory, PS-ID to file mapping, ESTIMATING run configuration (DEL-to-PKG mapping + per-package PRICE_SOURCES list + usage guidance), open issues, gaps. Full section contract: see INDEX_MD_CONTRACT.md.
10. Generate Scaffold/BOE_Scaffold.md. Per-deliverable table (DEL-ID, name, package, default basis, fallback policy, mixed-method flag, substance, cost drivers, roles), package cost ownership hints, SOW multi-mapping warnings. Marked DRAFT — requires human review.
11. QA + handoff artifacts. Produce Confidence_Summary.md, QA_Report.md with RUN_STATUS, Decision_Log.md, Assumptions_Log.md, Source_Index.md, Override_Log.csv (ENRICH only), Conflicts.csv (when needed), Publish_Manifest.md (always).

Phase BOE — method

1. Resolve tool root + create snapshot folder. Create EPREP_BOE_{LABEL}_{DATE}_{TIME}/.
2. Load approved scaffold. Read SCAFFOLD_PATH; diff against original scaffold if identifiable; log human modifications in Decision_Log.md.
3. Load decomposition + dependencies. Read DEPENDENCY_SOURCES (or AUTO for per-deliverable Dependencies.csv); build dependency DAG; detect cycles.
4. Derive tier sequencing. Assign tiers via topological layering (T0, T1, ...); identify sequential chains and parallelization opportunities; output Tier_Analysis.md.
5. Generate per-deliverable estimation plan. Tier assignment, basis/fallback/mixed-method from approved scaffold, substance and cost drivers, required price sources; package-level cost ownership rules.
6. Generate dependency-informed run sequence. Tiers in order + within-tier parallelism + sequential chains + gates.
7. Generate aggregation strategy. Rollups: deliverables to packages to project totals; add evaluation-weight views if EVALUATION_CRITERIA provided.
8. Generate missing/weak PRICE_SOURCES register. Identify and prioritize low-confidence items impacting the plan.
9. Compile assumptions and constraints log. Merge scaffold assumptions + BOE derivations with IDs and impact-if-wrong.
10. Compile full BASIS_OF_ESTIMATE.md. Follow canonical BOE structure if one exists; otherwise use the 10-section set: Purpose, Project Context, Estimation Scope, Estimation Strategy, Per-Deliverable Plan, Dependency-Informed Run Sequence, Missing/Weak PRICE_SOURCES, Aggregation Strategy, Assumptions Log, Document Control. Full section contract + handoff artifact contracts: see BOE_STRUCTURE.md.
11. QA + handoff artifacts. QA_Report.md with RUN_STATUS, Decision_Log.md, Assumptions_Log.md, Source_Index.md, Conflicts.csv (if cycles or contradictions), Publish_Manifest.md (always).

Outputs

SCAFFOLD snapshot

{EXECUTION_ROOT}/_EstimatePrep/EPREP_SCAFFOLD_{LABEL}_{DATE}_{TIME}/
  Run_Context.md
  PriceSources/
    [pricing library CSVs per RATE_SCOPE]
    INDEX.md
  Scaffold/
    BOE_Scaffold.md
    Package_Analysis.md
  Confidence_Summary.md
  QA_Report.md
  Decision_Log.md
  Assumptions_Log.md
  Source_Index.md
  Override_Log.csv          (ENRICH mode only)
  Conflicts.csv             (when needed)
  Publish_Manifest.md
  Publish_Package/          (optional; if EXPORT_BUNDLE=MANIFEST_AND_PACKAGE)

BOE snapshot

{EXECUTION_ROOT}/_EstimatePrep/EPREP_BOE_{LABEL}_{DATE}_{TIME}/
  Run_Context.md
  BASIS_OF_ESTIMATE.md
  Tier_Analysis.md
  QA_Report.md
  Decision_Log.md
  Assumptions_Log.md
  Source_Index.md
  Conflicts.csv             (when needed)
  Publish_Manifest.md
  Publish_Package/          (optional)

Non-negotiable constraints

• Write quarantine. Write ONLY under {EXECUTION_ROOT}/_EstimatePrep/. Never modify _PriceSources/, _Estimates/, deliverable folders, decomposition outputs, or dependency registers. Publishing to canonical locations is a separate, human-approved step handled by the invoker.
• Snapshots are immutable. Each run writes a new snapshot folder; never overwrite prior snapshots.
• Phase separation. SCAFFOLD and BOE are distinct invocations. A single run MUST NOT span both phases — the human gate between them is a non-negotiable decision point.
• Provenance tracking is mandatory. Every generated value carries provenance appropriate to its schema family: Basis + Confidence for Families 1-7; Source + Confidence for Family 8; Basis alone (no Confidence) for Family 9. Do not add columns that do not exist in the canonical schema for a given file.
• Parametric defaults are not human confirmation. Parametric values MUST be labeled PARAMETRIC with MEDIUM or LOW confidence, never HIGH. Only human-confirmed, vendor-quoted, or directly source-document-derived values earn HIGH.
• No silent overrides. When ENRICH mode overlays human data on a baseline, every override MUST be logged in Override_Log.csv (what changed, from what, to what, confidence change).
• Conflicts are surfaced. If two sources disagree or schema mismatches exist, produce Conflicts.csv; do not silently pick a winner.
• Dependencies are read-only inputs. Dependency registers inform tier sequencing in Phase BOE but are never modified by this skill.
• No recursive ingestion. Do not treat prior _EstimatePrep/ outputs as "market evidence" unless explicitly provided as PRIOR_SNAPSHOT / SCAFFOLD_PATH.
• CSV schema integrity. Every generated CSV MUST match the exact column names and order specified by its schema family (or by the canonical file discovered via AUTO_FROM_CANONICAL). No invented columns.

Human decision rights (must remain human)

This skill may propose, but MUST NOT decide:

• Accepting / issuing the BOE strategy and publishing it to canonical locations.
• Rulings on conflicts (two quotes disagree, scope overlaps, contradictory dependency assertions).
• Approving any override that would change a HIGH-confidence value.
• Scope boundary decisions (in/out; base vs option vs alternate).
• Any irreversible publication action (git commit/push; copy into _PriceSources/).

Human rulings SHOULD be recorded in the scaffold/BOE artifacts (or in a separate decision log maintained by the invoker).

QA expectations

See QA_CHECKS.md for the full invariant + quality gate set. Summary:

• RUN_STATUS declared in QA_Report.md (OK | WARNINGS | FAILED_INPUTS).
• Snapshot folder created for every run (including failures).
• No writes outside _EstimatePrep/.
• CSV schemas match the canonical Schema Annex exactly.
• Provenance complete per schema family.
• Publish_Manifest.md exists and references the run outputs.
• Phase-specific required artifacts present (see QA_CHECKS.md).

Companion documents

This skill's detailed contracts live in adjacent companion files. Consult these when implementing:

• SCHEMA_ANNEX.md — 9 canonical CSV schema families covering 18 files in _PriceSources/; Basis/Source/Confidence enum references; Override_Log.csv and Conflicts.csv schemas. Hardened against the on-disk canonical library as of 2026-02-18.
• BOE_STRUCTURE.md — BOE_Scaffold.md contract (Phase SCAFFOLD), BASIS_OF_ESTIMATE.md 10-section contract (Phase BOE), Run_Context.md minimum fields, Publish_Manifest.md handoff contract.
• INDEX_MD_CONTRACT.md — 7-section PriceSources/INDEX.md contract including the ESTIMATING run configuration block.