Skip to content

CHANGELOG.md

Latest commit

 

History

History
236 lines (194 loc) · 16.4 KB

CHANGELOG.md

File metadata and controls

236 lines (194 loc) · 16.4 KB
Raw

Changelog

All notable changes to the Sigint Market Intelligence Plugin will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.10.3] - 2026-05-01

Added

  • Adversarial falsification (/sigint:falsify): New skill performs adversarial assessment of research findings via web-only disconfirming search. Decomposes findings into atomic claims, generates hybrid template + LLM-counter-hypothesis queries, executes adversarial WebSearch/WebFetch, and assigns ordinal verdicts (falsified, weakened, survived, inconclusive)
  • Mandatory remediation loop: Falsified findings are quarantined with gate: "post-falsification"; weakened findings have confidence downgraded one level and disconfirming sources appended in-place; a falsification-followups.json queue feeds /sigint:issues with retraction/comment actions
  • falsification-analyst agent: New adversarial subagent that treats each finding as a hypothesis under test; web-only constraint, helpfulness-bias warning, one-round-per-session rule, and bounded-epistemics caveat
  • Phase 3.6 falsification gate: research-orchestrator invokes the falsify skill between post-merge codex review and progress rendering. Block-mode by default — any falsified verdict halts downstream phases. Configurable via sigint.config.json global.falsify.enabled
  • schemas/falsification-report.jq and schemas/falsification-followups.jq: New validators
  • docs/how-to/falsify-findings.md: How-to with verdict→remediation table, scopes, followups workflow, and troubleshooting
  • 6 evals for the falsify skill covering each verdict class, scope variants, advisory mode, missing-session error, and budget overflow handling

Changed

  • schemas/findings.jq: Added optional provenance.falsification_attempts[] and top-level requires_issue_followup fields (backward compatible)
  • schemas/quarantine.jq: Added post-falsification to the gate enum
  • schemas/state.jq: Added falsification to the lineage valid_action enum
  • agents/issue-architect.md v0.2.0: Reads falsification-followups.json if present and processes each item via open_issue / comment_issue / close_issue / annotate actions
  • agents/research-orchestrator.md v0.6.0: Added Phase 3.6 falsification gate with skip conditions, block/advisory mode handling, and progress logging

[0.9.0] - 2026-04-14

Added

  • Finer-grained delta tracking: UPDATED findings now carry a delta_detail object sub-classifying the change type (substantive, temporal, confidence_shift, source_refresh, metadata) and a newsworthiness signal (high/medium/low) for downstream consumers
  • Newsworthiness signal: Each UPDATED finding includes newsworthiness (high/medium/low) and newsworthiness_basis explaining the rating, enabling article generators and report writers to filter updates by significance
  • Delta detail fields: changed_fields (which mutable fields differ), change_category (dominant change type), summary_diff (what changed in the summary), confidence_change and trend_change (previous/current values)
  • Structured delta_from_previous in lineage: Lineage entries now include typed counts (new_count, updated_count, confirmed_count, potentially_removed_count, trend_reversal_count, newsworthy_count) and an update_breakdown by change category
  • Newsworthy Changes section in delta report: Delta report markdown now leads with a consolidated Newsworthy Changes section (all NEW + high-newsworthiness UPDATED + TREND_REVERSAL findings) and sub-categorizes Updated Findings by change type
  • schemas/delta-findings.jq: New schema validator for findings_delta_YYYY-MM-DD.json files, including delta_type and delta_detail validation
  • 5 new evals (IDs 9–13): Cover substantive vs temporal classification, confidence shift recording, newsworthy count in lineage, and Newsworthy Changes section in delta report

Changed

  • Delta Detection Protocol Step D.2.5: New sub-step for computing delta_detail on UPDATED findings with explicit classification guidance and mixed-type confidence handling
  • Delta report template: Restructured with Newsworthy Changes section, sub-categorized Updated Findings (Substantive, Confidence, Temporal, Metadata & Source), and breakdown counts in Summary
  • schemas/findings.jq: Added optional delta_type and delta_detail validation (backward compatible)
  • schemas/state.jq: Added optional structured delta_from_previous validation in lineage entries (backward compatible)

[0.6.0] - 2026-04-11

Removed

  • Atlatl MCP dependency: Removed all Atlatl blackboard, memory capture, memory recall, and memory enrichment references across 55 files (491 occurrences). JSON file persistence is now the sole data model.
  • Blackboard coordination: Replaced ephemeral blackboard key-value store with file-based state in reports/{topic_slug}/. Inter-agent coordination uses TaskUpdate and SendMessage exclusively.
  • auto_atlatl configuration field: Removed from schema validation, config examples, migration logic, and documentation.
  • atlatl_memory_id topic field: Removed from schema validation and lifecycle tracking.
  • 62 Atlatl-specific eval test cases: Removed from 7 eval files; all remaining evals validated.

Improved

  • skills/augment: Added --methodology as documented alias for --dimension argument
  • skills/report: Input validation, format-specific output, timeout handling, graceful cleanup, state.json validation
  • skills/issues: Input validation (--repo format), cowork fallback, categorized error responses, Atlatl regression guard eval
  • skills/migrate: Backup-before-write ordering, dual-format parsing (YAML + markdown sections), malformed JSON handling, schema validation fallback

Fixed

  • Topic lifecycle tracking: Research sessions now register in sigint.config.json topics throughout the lifecycle — /sigint:start registers with in_progress, orchestrator sets complete on finish, /sigint:augment and /sigint:update update dimensions and timestamps
  • Session index: /sigint:status and /sigint:resume --list now use sigint.config.json topics as primary session index instead of only globbing report directories
  • Schema validation: sigint-config.jq updated to validate both minimal (context-only) and lifecycle-managed topic entries with status, dimensions, created/updated timestamps, and findings count
  • Dimension-analyst reports directory: Orchestrator now passes explicit REPORTS_DIR and TOPIC_SLUG to each analyst spawn prompt; analysts use the path verbatim instead of deriving it from the topic title (fixes slug truncation causing findings to land in wrong directory)
  • Pre-review file validation (Phase 2.6): Orchestrator validates all expected findings files exist in the canonical reports directory before the codex review gate (not after), ensuring relocated files go through the blocking review. Recovery is fail-closed: single-match relocations only, refuses on ambiguous multiple candidates, never imports from sibling topic directories
  • Config write atomicity: Orchestrator Phase 4.1 now writes all topic completion fields (status, findings_count, dimensions) in a single jq call to prevent race conditions

0.5.0 - 2026-04-02

Added

  • Harness pattern: Research-orchestrator follows the Anthropic long-running agent harness pattern with research-progress.md for cross-session continuity
  • Codex review gates: 4 blocking gates at pipeline boundaries (post-findings, post-merge, post-report, post-issues) with quarantine mechanism for failures
  • Provenance enforcement: Every finding must carry a provenance record (claim, sources with URLs, derivation, confidence basis)
  • Delta detection protocol: Update mode classifies findings as NEW, UPDATED, CONFIRMED, POTENTIALLY_REMOVED, or TREND_REVERSAL with generated delta reports
  • Structured Data Protocol (protocols/STRUCTURED-DATA.md): All JSON operations use jq via Bash with mandatory schema validation after every write
  • 12 jq schema validators in schemas/ directory: state, findings, elicitation, methodology-plan, reflection, quarantine, merged-findings, report-metadata, issues, team-status, conflicts, sigint-config
  • Config Resolution Protocol (protocols/CONFIG-RESOLUTION.md): Defines config cascade for sigint.config.json v2.0 with project > global > hardcoded resolution
  • Migrate skill (skills/migrate/SKILL.md): Converts legacy sigint.local.md or .sigint.config.json v1.0 to sigint.config.json v2.0 with per-topic support, dry-run preview, and idempotent execution
  • trend_modeling dimension: 8th research dimension using three-valued logic scenario modeling (note: uses underscore, not hyphen)
  • 6 operational skills: start, update, augment, migrate, issues, report — thin launchers that delegate to specialized agents via swarm orchestration

Changed

  • All JSON file operations now use jq via Bash — Edit tool removed from all agents and skills per /refactor:xq structured data reliability patterns
  • Configuration format: Migrated from sigint.local.md YAML to sigint.config.json v2.0 JSON with per-topic overrides
  • Schema validation is mandatory: Write-then-validate pattern required after every JSON mutation with retry-and-correct (max 2 retries)
  • File persistence is default: All findings written to validated JSON files
  • Research-orchestrator upgraded to v0.5.0 with codex gates, provenance, delta detection, and harness pattern
  • Dimension-analyst now includes Bash in tools list for Structured Data Protocol compliance
  • Report-synthesizer now includes Bash in tools list for Structured Data Protocol compliance
  • Resume command follows harness initialization protocol (read progress files before acting)

Removed

  • Edit tool from all agents — replaced by Bash + jq for JSON operations
  • sigint.local.md as primary config — replaced by sigint.config.json v2.0 (legacy format supported via /sigint:migrate)

0.4.0 - 2026-03-20

Added

  • Swarm-orchestrated parallel research: New research-orchestrator agent coordinates multiple dimension-analysts running concurrently
  • Dimension-analyst agent: Generic research analyst parameterized by dimension (competitive, sizing, trends, customer, tech, financial, regulatory)
  • Source-chunker agent: RLM processor for large documents — partitions, spawns chunk analysts, synthesizes findings
  • Task-based coordination: Task system for inter-agent communication during research
  • Orchestration hints in all 9 skill SKILL.md files for team-based research participation
  • Live team status in /sigint:status showing dimension-analyst progress

Changed

  • Replaced monolithic market-researcher with research-orchestrator + dimension-analyst swarm (3→5 agents)
  • Replaced TodoWrite with TaskCreate/TaskUpdate in all commands
  • Replaced Task/subagent_type delegation with Agent tool in all commands
  • Report-synthesizer now reads dimension findings files in addition to state.json

Removed

  • market-researcher agent (decomposed into research-orchestrator + dimension-analyst)
  • Subcog integration (replaced by file-based state)
  • TodoWrite usage (replaced by TaskCreate/TaskUpdate)

0.3.7 - 2026-01-23

Fixed

  • Init command now creates project config with default template if missing (originally sigint.local.md, superseded by sigint.config.json v2.0 in 0.5.0)

0.3.0 - 2026-01-23

Added

  • Cascading configuration support: global defaults with project overrides (originally sigint.local.md, superseded by sigint.config.json v2.0 in 0.5.0)
  • Marketplace installation option (/plugins add sigint)
  • README.md generation for report folders:
    • Master index (./reports/README.md) listing all research topics with status and summaries
    • Topic index (./reports/<topic>/README.md) with:
      • Research query summary
      • Configuration/elicitation settings used
      • Links to all artifacts (reports, state, issues)
      • Key findings summary

Changed

  • Configuration resolution order: Project > Global > Built-in defaults
  • Updated installation documentation with three options (marketplace, local dev, manual)
  • market-researcher agent now generates README.md on initial research
  • report-synthesizer agent updates README.md when generating reports

Fixed

  • Broken links in executive-brief.md example (now correctly reference cross-skill examples)
  • Documentation clarity for configuration paths across all docs

0.2.1 - 2026-01-23

Fixed

  • Agent delegation: Commands now delegate to specialized agents instead of executing research/reports directly in the main REPL

0.2.0 - 2026-01-22

Changed

  • Reorganized plugin structure with comprehensive elicitation workflow
  • Enhanced /sigint:start command with structured research brief questions

Added

  • User documentation in docs/ directory:
    • quick-start.md - 5-minute getting started guide
    • workflow-guide.md - Complete research lifecycle documentation
    • troubleshooting.md - Common issues and solutions
  • GitHub social preview image (.github/social-preview.svg)
  • README infographic showing research workflow
  • Quick Start section in README with example workflow
  • CONTRIBUTING.md - Contribution guidelines and development setup
  • SECURITY.md - Security policy and vulnerability reporting
  • .gitattributes - Line ending normalization

Fixed

  • License field in plugin.json corrected from Apache-2.0 to MIT
  • LICENSE file year updated to 2026

0.1.0 - 2026-01-22

Added

Agents

  • market-researcher - Comprehensive market research, competitive intelligence, and strategic insights
  • report-synthesizer - Transforms raw research findings into polished, executive-ready documents
  • issue-architect - Converts research findings into actionable GitHub issues

Commands

  • /sigint:start - Begin a new market research session with comprehensive scoping and elicitation
  • /sigint:augment - Deep-dive into a specific research area
  • /sigint:init - Initialize plugin configuration for a project
  • /sigint:issues - Generate GitHub issues from research findings
  • /sigint:report - Generate a comprehensive report from research
  • /sigint:resume - Resume an existing research session
  • /sigint:status - Show current research session state and progress
  • /sigint:update - Refresh data and findings for existing research

Skills

  • Competitive Analysis - Porter's Five Forces, competitive matrices
  • Market Sizing - Top-down and bottom-up methodologies
  • Trend Forecasting - Three-valued logic trend models
  • Customer Research - Persona development and journey mapping
  • Technology Assessment - Build vs. buy analysis frameworks
  • Financial Analysis - Unit economics and SaaS metrics
  • Regulatory Review - Compliance and regulatory impact assessment
  • Report Writing - Executive briefs and comprehensive reports
  • Data Collection - Primary and secondary research techniques

Templates

  • Market sizing report template
  • Competitive analysis report template
  • Trend analysis report template
  • Technology assessment report template

Reference Materials

  • Porter's Five Forces framework guide
  • Competitive matrix templates
  • Market sizing methodologies
  • Three-valued logic documentation
  • SaaS metrics reference
  • Regulatory compliance checklists

Examples

  • Competitive analysis report example
  • Market sizing report example
  • Trend analysis report example
  • Technology assessment example
  • Financial analysis example
  • Customer journey map example
  • Executive brief example
  • Full comprehensive report example

Documentation

  • README with plugin overview and quick start guide
  • Research methodology citations (arXiv:2601.10768)
  • MIT license