Total Quests
58
QUEST INTELLIGENCE
Total Quests
58
Finished
56
In Progress
0
Blocked
0
Abandoned
2
Current normalized state across all quests
Monthly trend using each quest's final/current status
58 quests represented
Improve the standalone sharpen skill so its questions are grounded in repo evidence when local implementation facts matter. Context: - The canonical skill is `.skills/sharpen/SKILL.md`. - `.agents/...
$quest implement ideas/2026-05-14-pr-shepherd-operational-intake.md
Planned YAML for `.github/workflows/security.yml` (`workflow-guard` job): ```yaml workflow-guard: runs-on: ubuntu-latest steps: - name: Checkout repository uses: actions/checkout@v4
Completed successfully.
Fixed issue #110 so repeated same-day installer upgrades no longer fail when the default `quest-update-YYYYMMDD` branch already exists locally.
Completed successfully.
Completed successfully.
Impact: - Adds one canonical machine-readable artifact (`/tmp/deep_ci_context_manifest.json`) per run so selection/chunking/omission decisions are inspectable and deterministic. - Keeps prompt behavior stable while moving Deep CI decision-making into a single source of truth consumed by rendering...
Completed successfully.
Phase 2 of review-intelligence-canonical shipped. pr-shepherd now normalizes incoming review items into the canonical Phase 1 contract, decides per-finding action via the shared review-decisions policy, batches actionable fixes by write_scope + validation scope, runs the smallest falsifying checks (Level 0/1/2), and pushes once per validated batch with explicit bounded stop conditions.
Implement Phase 1 of review-intelligence-canonical: normalize review
Dashboard quest detail pages now include the brief and celebration context, and archived journal pages were backfilled safely.
Add two narrow, artifact-backed carry-over sections to Quest celebration/journal output so Phase 1 review intelligence becomes visibly useful without changing the broader celebration design. The implementation should extend `QuestData` with structured carry-over fields, populate them from the exi...
Turned the Claude insights evaluation into sanity-checked technical idea documents with clear prioritization.
Consolidated overlapping Quest prompt-surface proposals into one canonical instruction architecture document.
Consolidate the Quest routing and feedback-intent ideas into one canonical delegation proposal. Use `ideas/2026-04-13-feedback-aware-delegation-keywords.md` as the base and absorb the useful compan...
Reviewed `.ws/caveman`, captured the useful learnings, and explicitly decided not to take implementation action now. The caveman style itself was rejected for readability reasons; the only keepers were a few general patterns around evaluation rigor, safety escape hatches, and guarded compression tooling.
Multi-cleanup quest. Continuing on our existing branch. fix/quest-startup-outside-repo. In ideas, we have several things that should be archived. Don't assume that they are done or not done based o...
Extract the embedded Python from .github/workflows/codex-ci-review.yml into a standalone script at .github/scripts/codex_review.py.
User selection: full quest.
Forensic sweep across stale quests, missing journals, broken automation, and a Codex sandbox permission footgun.
Installer handles Codex MCP setup; bridge timeout raised to 30 minutes.
Documentation cut by more than half without losing substance.
Closed the state-transition footgun and enforced the mandatory presentation gate.
Artifact preparation is now explicit and happens before role execution, not during.
Codex MCP setup docs overhauled, allowlist cleaned up, model dispatch reads from config instead of hardcoding.
New skill for delegating tasks to Codex via MCP.
First-class Codex-led Claude runtime path for Quest orchestration.
Retired from `ideas/` on 2026-03-09.
Status: complete
Status: complete
Bridged two isolated data readers (celebrate `quest_data.py` and dashboard `loaders.py`) so that `/celebrate` can replay celebrations from archived journal entries. Added an 8-tier candid quality scale where smooth quests get Diamond and rough ones get Tin or Cardboard. Dashboard cards now show quality tier badges with hover tooltips, agent model "Cast" lines, and test counts.
Added a practical inline comment playbook to PR shepherd guidance so review replies are kind, precise, and actionable.
Reworked quest celebration system with deep artifact reading, block-letter titles, dynamic achievements, impact metrics, quality scores, and cinematic movie-style credits.
Implemented Quest Completion Animation System with 4 animation styles, 38 passing tests, and integration into quest workflow.
Created comprehensive documentation (`docs/guides/opencode-model-suitability.md`) mapping all 32 OpenCode models to Quest orchestration roles. The guide includes:
Added `.github/workflows/pr-body-gate.yml` — a GitHub Actions workflow that validates PR bodies contain four required headings: - `## Summary` - `## Changes` - `## Validation` - `## Notes`
Relocated six Quest role files from `.ai/roles/` to `.skills/quest/agents/`, then updated runtime references, validators, metadata, and docs to match the new ownership model.
Implemented `scripts/quest_validate-quest-state.sh` — the first system-enforced correctness check for Quest phase transitions. The shell script validates state.json integrity, phase transition legality, artifact prerequisites, semantic content checks (arbiter verdict, review outcomes via jq on handoff JSON), and iteration bounds before every phase transition.
Implemented the handoff.json structured file pattern to close the remaining context leaks in the Quest orchestrator. Every agent now writes a tiny JSON file (~200 bytes) alongside its artifacts, and the orchestrator reads this file for routing decisions instead of processing full agent responses.
Restructured the Quest Dashboard layout to match the target executive "Quest Intelligence" design from PR #21. The dashboard now features the "QUEST INTELLIGENCE" branding eyebrow, "Quest Portfolio Dashboard" title, 5 KPI cards (Total, Finished, In Progress, Blocked, Abandoned), side-by-side chart panels (Status Distribution doughnut + Final Status Over Time line chart), and a unified Quest Portfolio section with redesigned cards showing full pitch text and labeled metadata.
Fixed HTML attribute injection (XSS) vulnerability in the Quest Dashboard's URL rendering. Added `_sanitize_url()` function that validates scheme (HTTPS-only), validates GitHub URL pattern via regex, and applies HTML attribute escaping before any URL is interpolated into `href="..."` attributes. Also escaped fallback relative paths, removed stderr side effect from loaders, and unified the `github_url` missing-value convention.
Restored visual features from the PR #21 prototype that were cut during the dashboard-v2 implementation. Added ambient CSS glows (green, blue, amber blurred circles behind content), interactive Chart.js doughnut and stacked area charts, and gradient enhancements on cards and section headers. Chart.js 4.4.7 is vendored and inlined at render time — the dashboard remains a single self-contained HTML file with no external dependencies. Graceful degradation ensures the dashboard still works if the vendor file is missing.
Built the Quest Dashboard: a self-contained Python package that generates a static HTML dashboard from quest journal entries and active quest state files. Three status sections (Finished, In Progress, Abandoned) with dark navy theme, elevator pitch cards, journal/PR links, and muted metadata. Single-file HTML output with no external dependencies.
Added a GitHub Actions workflow to run the Python test suite (pytest) on every push to main and PR targeting main. This ensures the 36 Quest Dashboard unit and integration tests are enforced in CI, preventing regressions from slipping through undetected.
First attempt at building the Quest Dashboard. Plan was approved by arbiter after one iteration, but the build phase was interrupted when models were switched from Sonnet to Opus. Rather than resuming, a new quest (dashboard-v2) was created using this quest's approved plan as input. Dashboard-v2 completed successfully.
Ported the Codex CI code review workflow from candid_talent_edge (PR #164) to the Quest framework repository. Added a GitHub Actions workflow that triggers OpenAI Codex as an automated code reviewer when a PR transitions from draft to ready-for-review, a Quest-adapted CI code reviewer skill, and registered the new skill in the catalog and manifest.
Implemented Phase 2 of the Quest architecture evolution: the "thin orchestrator" principle. The Quest orchestrator now passes file paths to subagents instead of embedding content, reducing context growth from O(n*content_size) to O(n*one_line).
Analyzed how Quest should organize, manage, and distribute skills. Researched community best practices (Agent Skills standard, plugin marketplaces, distribution patterns). Produced strategic recommendations.
Fixed handoff contract inconsistencies across Quest role files and workflow. Standardized all 6 role files on `---HANDOFF---` text format with STATUS/ARTIFACTS/NEXT/SUMMARY fields, removed "Context Is In Your Prompt" contradictions, and made workflow prompts explicitly request handoff format.
Decomposed the monolithic `.skills/quest/SKILL.md` (635 lines) into a slim routing entry point (73 lines) plus three delegation files under `.skills/quest/delegation/`. The delegation pattern structurally enforces intake quality — vague input routes to a dedicated questioning phase before planning begins, while detailed input goes directly to the workflow.
Comprehensive research exploration mapping 11 distinct caching strategies applicable to the Quest AI agent system. The quest produced two educational research documents and a quest index README — no code changes, as the research itself was the deliverable.
Implemented automatic weekly update checking for Quest. After a quest completes, the system checks if updates are available (respecting a configurable interval) and prompts the user to update if a newer version exists upstream.
Planned a dual-planning-track feature for `/quest` where users could choose between "Quest" (fast, single plan) and "Quest Council" (rigorous, two competing plans merged by arbiter). Plan was approved after 2 iterations but was never built.
First-ever quest run on the extracted repository. Validated that the Quest blueprint was correctly extracted from the source repo, created the `ideas/` directory for tracking future work, and updated the README with public domain messaging.
Enhanced the quest orchestration to present plans interactively rather than dumping the full plan at once. Users now see a brief summary first, then can opt into a phase-by-phase walkthrough with the ability to request changes at each step.
Created a single unified installer script (`scripts/quest_installer.sh`) that installs and updates Quest in any repository.
Implemented GitHub Actions CI and local pre-commit hooks that validate quest-related artifacts to prevent broken configurations from being committed.