Detailed PR analysis with signals and checklist. Use analyze_pr for deep analysis, coach_pr for quick pass/fail.

Returns: tier (low_risk/medium_risk/high_risk), signals (positive/negative/neutral), checklist (P0/P1/P2 items), and recommended_action.

Use this before submitting a PR to a large open-source repo to check if your contribution strategy is sound. Combines repo profile data, anti-pattern matching, and success pattern analysis.

Args: title: PR title (e.g. "feat: add new search endpoint") repo: Target repo in org/name format (e.g. "encode/httpx") body: PR body/description text description: Extended description (optional) author: PR author username author_association: Author's association with repo (NONE/CONTRIBUTOR/COLLABORATOR/MEMBER/OWNER) labels: PR labels (e.g. ["bug", "documentation"]) star_count: Repository star count (>= 0; 0 = unknown) repo_merge_rate: Repository's external PR merge rate 0.0-1.0 (0 = unknown) mergeable: Mergeable state (MERGEABLE/CONFLICTING/UNKNOWN) diff_stat: git diff --stat output — populates impact/review and is the authoritative PR-size signal (null when omitted)

Returns: dict with keys: tier, signals, checklist, recommended_action, repo, title, impact, review (impact/review null without diff_stat)

Example: analyze_pr("fix: timeout in connection pool", "encode/httpx", "Fixes #123") → {"tier": "medium_risk", "signals": {...}, "checklist": [...], "recommended_action": "..."}

Agent PR Dojo: quick pass/fail gate. Use coach_pr for go/no-go decision, analyze_pr for detailed analysis.

Use this as a final gate before opening a PR. Returns pass=true if safe to submit, pass=false if there are blocking issues to fix first.

Combines repo profile analysis, anti-pattern matching, and maintainer policy checks into a single go/no-go decision.

Args: title: PR title repo: Target repo (org/name) body: PR body/description description: Extended description author: PR author username author_association: Author's association (NONE/CONTRIBUTOR/etc.) labels: PR labels star_count: Repo star count (>= 0) repo_merge_rate: External PR merge rate (0.0-1.0) mergeable: Mergeable state (MERGEABLE/CONFLICTING/UNKNOWN) diff_stat: git diff --stat output (populates impact/review)

Returns: dict with pass (bool), tier, signals, checklist, recommended_action

Policy-aware PR screening.

Reads docs/policies/-.md and checks the PR against hard/soft rules.

返回: verdict: pass / needs_preflight / warn / reject policy_loaded: bool violations: list of {rule, severity, evidence} generic_checks: list (only when needs_preflight) recommended_action: str (always)

Example: triage_pr("docs: typo", "Ikalus1988/MisakaNet", "fix typo", "docs/faq.md | 3 ++-") → {verdict: "pass", policy_loaded: true, ...}

triage_pr("docs: add installation", "pallets/flask", ...)
→ {verdict: "needs_preflight", generic_checks: [...6 items...], ...}

Get repository profile with maintainer guidelines, AI policy, and merge rate.

Returns 17 agent_guidelines fields including ai_policy, maintainer_vibe, external_merge_rate, close_keywords, and more. Use this to understand a repo's contribution culture before submitting a PR.

Args: repo: Repository in org/name format (e.g. "encode/httpx")

Returns: dict with repo metadata and agent_guidelines, or error if not found

List open PR case-study records in the local knowledge base (issue #74).

NOTE: these are local case-study folders with final_status=open — NOT live GitHub PRs. For live GitHub PR health use status_prs.

Args: repo: filter by target repo (org/name, e.g. "encode/httpx"); "" = all author: filter by PR author username; "" = all

Returns: list of {repo, pr_number, pr_url, author, folder} for each matching open case study

Get a specific PR case study with full details and rounds.

Returns the complete case study including frontmatter, body text, and round-by-round interaction history. Useful for learning from past PR experiences.

Args: repo: Repository in org/name format pr_number: PR number (integer)

Returns: dict with frontmatter, body, path, or error if not found

Search anti-patterns and success-patterns by keyword.

Use this to find relevant patterns before submitting a PR. Anti-patterns show what NOT to do, success-patterns show what works.

Args: query: Search keyword (e.g. "duplicate PR", "missing tests", "out of scope", "breaking change") pattern_type: Filter type — "all" (default), "anti-pattern", or "success-pattern" limit: Max results to return (default 10)

Returns: list of dicts with keys: key, title, symptom, fix_action, source_pr, type, file

Example: search_patterns("duplicate PR") → finds anti-patterns about duplicate submissions search_patterns("timeout", "anti-pattern") → finds timeout-related failure patterns

返回支持的 schema 版本和枚举值。

Check health of in-flight PRs (outbound PR heartbeat).

Classifies open PRs into 9 statuses: NEEDS_REBASE, CI_FAILING, STALE_REVIEW, CHANGES_REQUESTED, STALE_NO_REVIEW, BLOCKED, CLEAN, UNKNOWN, WAITING. Suggests actions for each.

Args: author: GitHub username to check PRs for repo: Repository to check (org/name) stale_days: Days without update to consider stale (default: profile or 14) save_snapshot: If True, persist the result to data/status-snapshots/ for transition detection across runs. Default False because MCP callers (Cursor, Claude Code) may poll frequently and would otherwise flood the snapshot directory.

Returns: dict with prs, ignored, summary, actions, transitions

Get profile writeback suggestions based on PR status analysis.

Analyzes PR health data and suggests profile updates (e.g., stale_days_threshold, response_time). Returns suggestions with evidence, source, and confidence score.

Args: author: GitHub username mode: 'suggest' (all suggestions) or 'auto' (confidence >= 0.8 only)

Returns: list of {field, value, repo, evidence, source, confidence, private}

Maintainer-facing action decision for a single PR (v1.5.1).

Output answers: 'What should I do with this PR right now?' Maps analyze_pr signals to 5 actions: READY_FOR_REVIEW, WAIT_FOR_AUTHOR, CLOSE_DUPLICATE, CLOSE_STALE_OR_RISKY, HOLD_MAINTAINER_DECISION.

Phase 5.1: Adds impact (files/lines/breaking/security), review (complexity/estimated_minutes), and author_info fields.

Args: title: PR title repo: Target repo (org/name) body: PR body description: Extended description author: PR author username author_association: Author's association (NONE/CONTRIBUTOR/etc.) labels: PR labels star_count: Repo star count repo_merge_rate: External PR merge rate 0.0-1.0 diff_stat: Diff stat string (e.g. "3 files changed, 10 insertions(+), 5 deletions(-)")

Returns: dict with keys: persona, repo, title, author, action, reason, blocking_signals, next_step, review_ready, impact, review, author_info, context.

Contributor-facing action decision for a single PR (v1.6.0).

Output answers: 'Is my PR ready to submit? What should I fix?' Maps analyze_pr signals to 5 actions: READY_TO_SUBMIT, FIX_BEFORE_SUBMIT, NEEDS_DISCUSSION, IMPROVE_CHANCE, ASK_MAINTAINER.

Args: title: PR title repo: Target repo (org/name) body: PR body description: Extended description author: PR author username author_association: Author's association (NONE/CONTRIBUTOR/etc.) labels: PR labels star_count: Repo star count repo_merge_rate: External PR merge rate 0.0-1.0 diff_stat: Diff stat string

Returns: dict with keys: persona, repo, title, author, action, action_label, action_icon, reason, next_step, checklist, confidence, merge_probability, impact, review, author_info, context.

Build maintainer review queue digest from a list of PRs (v1.5.0).

Groups PRs by maintainer action (5 classes) and produces a digest suitable for Actions summary or git-trackable markdown file.

Args: prs: List of {repo, number, title, body, author, labels, star_count, repo_merge_rate} prs_file: Alternative — path to JSON file containing the PR list

Returns: dict with keys: generated_at, total, summary (action -> count), results (per-PR maintainer_view), digest_markdown

Self-test the pr-genius install (issue #70).

Run this before trusting analyze_pr output. Reports Python/package versions, knowledge-base readability and coverage (anti-patterns, success-patterns, profiles, case studies, policies), gh CLI availability, MCP tool registration, a sample analyze against a known profile, and data-coverage warnings.

Returns: dict with keys: ok, prgenius_version, python_version, platform, knowledge_base, gh, mcp, sample, warnings