All published releases of SLOPE. See the full history on GitHub.
Full Changelog: https://github.com/srbryers/slope/compare/v1.48.1...v1.49.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.48.0...v1.48.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.47.2...v1.48.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.47.1...v1.47.2
Full Changelog: https://github.com/srbryers/slope/compare/v1.47.0...v1.47.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.46.1...v1.47.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.46.0...v1.46.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.45.1...v1.46.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.45.0...v1.45.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.44.1...v1.45.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.44.0...v1.44.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.43.1...v1.44.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.43.0...v1.43.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.42.0...v1.43.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.41.0...v1.42.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.40.0...v1.41.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.39.0...v1.40.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.38.0...v1.39.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.37.2...v1.38.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.37.1...v1.37.2
Full Changelog: https://github.com/srbryers/slope/compare/v1.37.0...v1.37.1
Full Changelog: https://github.com/srbryers/slope/compare/v1.36.0...v1.37.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.35.0...v1.36.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.34.0...v1.35.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.33.0...v1.34.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.32.0...v1.33.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.31.2...v1.32.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.31.1...v1.31.2
Full Changelog: https://github.com/srbryers/slope/compare/v1.31.0...v1.31.1
Guards can now return rich Suggestion objects with titled options, priority levels, and adapter-specific formatting. This replaces raw text blocks with interactive guidance.
id, title, context, options[], requiresDecision, prioritycritical forces deny, high/normal inject context onlyslope briefing at session start (fires once per session)slope phase status — Show current phase cleanup gate statusslope phase complete <N> — Mark a phase as completeslope phase reset <N> — Reset phase gatessession-state.ts — Atomic session state for guard dedup (tmp + rename pattern)phase-cleanup.ts — Phase gate tracking (5 gates: scorecards, handicap, map, findings, regression){} (was silently swallowing output)Run slope hook add --level=full after upgrading to install the 5 new guards and pick up the suggestion formatting.
Full Changelog: https://github.com/srbryers/slope/compare/v1.30.0...v1.31.0
computeHandicapTrend()computeVelocity()computeGuardMetrics() analytics/api/guard-metrics route, /api/data now includes handicapTrend and velocityslope stats export now includes handicap_trend and velocity fieldsmodel param now blocked with message to resubmit with model: "haiku". Previously silently inherited Opus (~18x cost). Fixes #173matcher: "Task") before adding fresh ones. Non-SLOPE hooks preserved.After upgrading, run slope hook add --level=full to:
"Task" → "Agent")This is required for the subagent cost-control fix to take effect.
Full Changelog: https://github.com/srbryers/slope/compare/v1.29.0...v1.30.0
Full Changelog: https://github.com/srbryers/slope/compare/v1.28.0...v1.29.0
Semantic code search that returns relevant snippets (~3k tokens) instead of requiring 60k-token explore subagents. Uses embedding index when available, falls back to grep -F with a setup tip.
top param bounded (1-50) to prevent abuseformat: 'paths' | 'snippets' — no full-file mode (use Read for that)slope map now shows full function signatures from SLOPE_REGISTRY (116 entries) instead of bare export names. Type-only exports dropped (discoverable via search({ module: 'types' })).
src/mcp/index.ts — context_search tool, config param on createSlopeToolsServer()src/mcp/registry.ts — registry entries for discoverabilitysrc/cli/commands/map.ts — signature lookup, type line removal, empty header cleanuptests/mcp/index-src.test.ts — registration, config, grep fallback integration testsmax_turns references from subagent-gate guard documentationAgent matcher instead of TaskIf your .claude/settings.json has a Task matcher for the subagent-gate hook, re-run:
npx slope hook add --level=full
This regenerates the hooks config with the correct Agent matcher.
guidance.mapStaleWarnAt/mapStaleBlockAt.slope auto-card or slope validate.roadmap.json automatically updates to "complete" when slope validate succeeds. Phase status also updates when all sprints complete.validateRoadmap() now warns about status mismatches (scorecard exists but status isn't "complete") and phantom sprints (marked complete but no scorecard)..slope/deferred-findings.json for tracking cross-sprint review findings.slope review defer --from=N --to=M --severity=medium --description="..."slope review deferred [--sprint=N] [--status=open]slope review resolve --id=<uuid>slope briefing output for the target sprint./start-sprint — Pre-sprint setup: briefing, branch creation, sprint state init, prior scorecard verification./post-sprint — Post-sprint: scorecard creation, validation, review generation, common-issues distillation./review-pr — Structured PR review with slope review recommend, finding tracking, and scorecard amendment.Installed by slope init --claude-code to .claude/commands/.
cwd to loadConfig() for correct custom threshold resolutionSLOPE_BIN_PREAMBLE set _SLOPE_BIN="slope" (bare name), causing the slope() shell function to call itself infinitely (SIGSEGV exit 139). Now resolves the full binary path via $(command -v slope).session_id, cwd, and hook_event_name when Claude Code omits them from stdin.Task to Agent to match current Claude Code tool name.Full Changelog: https://github.com/srbryers/slope/compare/v1.26.0...v1.26.1
slope doctor now detects when config.slopeVersion falls behind package.json and auto-fixes itslope doctor --fixslope guard recommend subcommand analyzes your workflow (sprint-workflow, monorepo, multi-session, flows) and suggests relevant missing guardsSLOPE_BIN_PREAMBLE (project → global → npx fallback)Full Changelog: https://github.com/srbryers/slope/compare/v1.25.5...v1.26.0
4 new MCP tools for structured manual testing sessions:
testing_session_start — Creates an isolated git worktree, returns setup steps from configtesting_session_finding — Records bugs/observations with severity during testingtesting_session_end — Ends session, returns summary by severity, cleans up worktreetesting_session_status — Shows active session info and findingsNew testing section in .slope/config.json:
{
"testing": {
"setup_steps": ["cd {worktreeRoot} && pnpm install"],
"teardown_steps": ["pnpm test"]
}
}
Supports {projectRoot} and {worktreeRoot} template variables.
search({ module: 'testing' }) discoveryslope doctor command (#123).slope/ to .gitignore during slope init (#122)Sprint completion guard — enforces post-implementation gates (Gates 4.5–7) before PR creation or session end.
.slope/sprint-state.json — tracks sprint lifecycle phase + 5 gate progression (tests, code_review, architect_review, scorecard, review_md)sprint-completion guard (3 hook points):gh pr create when gates are incompletejest/vitest/bun test exit 0) and marks tests gateslope sprint CLI — start --number=N, gate <name>, status, resetslope validate marks scorecard gate, slope review marks review_md gateFull Changelog: https://github.com/srbryers/slope/compare/v1.19.1...v1.20.0
findPlanContent(cwd) only searched {cwd}/.claude/plans/, but Claude Code writes plan files to ~/.claude/plans/ (global user directory). The guard never found any plan files when triggered through the hook pipeline.
tool_input.file_path instead of relying on directory scan~/.claude/plans/ as fallback search path, with deduplication when cwd equals homedirPostToolUse:Write when a plan file (.claude/plans/*.md) is writtenAskUserQuestion for review tier selectionfindPlanContent, countTickets, countPackageRefs, extractFilePatterns, extractTicketInfo into src/cli/guards/plan-analysis.tsslope review findings add, scorecard amendment via slope review amend, and pattern distillation via slope distill --autoAdded --staging flag to slope loop continuous that consolidates all sprint PRs in a batch into a single umbrella PR, reducing review burden and keeping main clean.
Usage:
slope loop continuous --staging --max=6
How it works:
loop/batch-<sprintId> staging branch from origin/mainNew module: src/cli/loop/staging.ts
initStagingBranch() — create/reuse staging branchcreateUmbrellaPr() — batch summary PR with sprint results tablecleanupStagingBranch() — safe delete after mergeChanges across the loop infrastructure:
worktree.ts — optional baseBranch param for branching from stagingpr-lifecycle.ts — baseBranch param for createPr/hasCommitsAhead, isStagingMerge option for autoMerge (skips file-count gate)executor.ts — threads staging branch through worktree/PR/merge, fetches staging ref after each sprint mergecontinuous.ts — staging init before loop, umbrella PR + cleanup afterFully backwards compatible — without --staging, behavior is unchanged.
When scorecard analysis produces empty backlogs (all hotspots resolved, 100% success rates), the analyze pipeline now falls back to structured planned sprints from roadmap.json. Scorecard-driven sprints always take priority. Caps at 3 sprints per regeneration cycle to force regression re-checks between batches.
planned array in roadmap.json with 6 Phase 10 sprint definitionsPlannedTicket/PlannedSprint types in loop types'roadmap' strategy routed to API model in model-selectorTwo-phase ticket execution — planner generates concrete file-level execution plans (target files, actions, test files), executor passes them as structured context to Aider.
slope prep injection with structured plan promptsRewrites the slope-loop/*.sh shell scripts into a first-class slope loop CLI command with type safety, testability, and security hardening.
| Subcommand | Description |
|---|---|
slope loop status |
Show loop progress, next sprint, config |
slope loop config |
Loop configuration management |
slope loop run |
Single sprint execution |
slope loop continuous |
Multi-sprint loop with backlog regen |
slope loop parallel |
Dual-sprint parallel with overlap detection |
slope loop results |
Format/display sprint results |
slope loop analyze |
Mine scorecards → generate backlog |
slope loop models |
Model selection analytics |
slope loop guide |
SKILL.md word count, hazard check |
slope loop clean |
Cleanup stale artifacts |
src/cli/loop/ with clean module separation.slope/loop.config.json → defaultsexecFileSync with array args (no string interpolation)git reset --hardkill(-pid)) for clean shutdownignore to prevent interactive hangsLoopConfig (16 fields with env var mapping)BacklogFile, BacklogSprint, BacklogTicketSprintResult, TicketResultAiderOutcome enum for spawn error detection.claude/worktrees/ to .gitignore so worktree directories don't show as untrackedWhen running parallel Claude Code sessions with worktrees, the stop-check guard was blocking sessions from exiting because it always checked the main checkout's status — seeing dirty files from other sessions as uncommitted work.
68 commits since v1.13.2 spanning sprints S42–S60. This release brings multi-sprint initiative orchestration, a self-development loop that runs sprints autonomously, and the infrastructure to support it.
pending → planning → plan_review → executing → scoring → pr_review → completeslope initiative create|status|next|advance|review|checklistslope-loop/run.sh — single sprint runner with tiered model routing (local Ollama + API fallback)slope-loop/continuous.sh — loop runner with backlog auto-regenerationslope-loop/parallel.sh — parallel runner with module overlap detection (git worktrees)slope-loop/analyze-scorecards.ts — mines scorecard data for backlog generationslope-loop/model-selector.ts — data-driven model tier recommendationsslope-loop/dashboard.ts — static HTML dashboard (handicap, model rates, costs, convergence)slope-loop-guide/SKILL.md) for sprint executionslope index CLI for building/refreshing the indexslope context --ticket=<id> for retrieving relevant code snippetsslope prep <ticket> — generates execution plans per ticketslope enrich <backlog> — enriches backlog tickets with file context and token estimatesgit reset --hard + git clean -fdguards.test.ts false positive from stop-check during loop runs--auto-test disabled for local modelspgrep replaces fragile ps aux | grep for loop detectiongtimeout support)slope init now writes npx -y mcp-slope-tools instead of the broken npx @slope-dev/slope/mcp subpath exportvalidateRoadmap no longer rejects valid cross-sprint depends_on references (e.g., S3-1 → S2-1)validateRoadmap, computeCriticalPath, findParallelOpportunities) now reference each other in MCP search resultsnpm install -g @slope-dev/[email protected]
Re-run slope init in existing projects to fix MCP configs.
??) now produce a non-blocking warning instead of blocking. This fixes an infinite loop when orphaned files from previous sessions are present but shouldn't be committed.Integrates implementation review rounds into SLOPE's scoring framework. Reviews are recommended based on sprint characteristics, findings are tracked as structured data, and scorecards are amended with review-sourced hazards that affect the final score.
slope review recommend — Recommends review types (architect, code, ml-engineer, security, ux) based on sprint metadata (ticket count, slope, file patterns)slope review findings add/list/clear — Manages review findings as structured data in .slope/review-findings.jsonslope review amend [--sprint=N] — Injects review findings as hazards into scorecard shots and recalculates score via buildScorecard()needs-amend guard state — Next-action guard suggests amend when findings exist but scorecard hasn't been amendedrecommendReviews(input) — Pure function returning ReviewRecommendation[]findingToHazard(finding) — Converts ReviewFinding → HazardHit using REVIEW_TYPE_HAZARD_MAPamendScorecardWithFindings(scorecard, findings) — Idempotent scorecard amendment with deduplicationReviewType: 'architect' | 'code' | 'ml-engineer' | 'security' | 'ux'ReviewFinding, ReviewRecommendation, AmendResultREVIEW_TYPE_HAZARD_MAP: architect→bunker, code→rough, ml-engineer→rough, security→water, ux→treesSeven SLOPE gaps filled to power the CaddyStack MVP:
SlopeStore implementation with multi-tenancy (project_id), transaction-scoped advisory lock migrations, and JSONB-aware row mappingFileProjectRegistry for managing multiple project configsinitFromInterview() + slope init --interactive for guided project setupGitHubClient using native fetch() with typed errors, auto-pagination, and rate-limit handlingcheck_run/workflow_run handlersaggregateStandups() for multi-agent team standups with conflict detectionpnpm test:pg script for local dev with Dockernpm install @slope-dev/[email protected]
PostgreSQL support (optional):
npm install pg