Skip to main content

Alfred — Agent Runbook: workflow routing, SOP checklists, and document management for AI agents

Project description

Alfred

Alfred

Agent Runbook

Workflow routing, SOP checklists, and document management for AI agents and humans.

PyPI Tests Python 3.10+


What is Alfred?

Alfred is a CLI-based agent runbook (af) that manages SOPs, workflows, and structured documents across three layers (PKG, USR, PRJ). It provides:

  • NEW in v1.10.0 — Agent-editable helpers and skill documents: af agent call/run, af skill list/read, and af plan --with-skills
  • v1.9.1GitHub App PR Review Bot Loop (COR-1615) for Codex Connector / Copilot PR review loops; v1.9.0 added Council Review (COR-1613) and Diagnose Feedback Loop (COR-1503)
  • Workflow Routingaf guide tells AI agents which SOP to follow for any task
  • Workflow Checklistsaf plan generates step-by-step checklists from SOPs. With --task "<description>" auto-composes the SOP set from tags; --todo flattens into a unified checklist; --graph renders ASCII + Mermaid flowcharts with intra-SOP loops and gates; --with-skills recommends matching skill docs
  • Agent Helpers & Skillsaf agent runs explicitly gated local Python helpers/scripts, while af skill discovers and reads reusable REF/SOP skill documents without executing code
  • Document Validationaf validate enforces metadata format, status values, and section structure
  • Document Formattingaf fmt normalizes metadata order, whitespace, and table alignment to canonical style
  • File Path Lookupaf where prints the absolute filesystem path of any document by identifier
  • Document Lifecycle — Create, read, update, search, and index documents with consistent naming
  • Tags Metadata — Optional Tags: field, exposed via Document.tags and filterable with af list --tag
  • JSON Output--json flag on guide/plan/search/validate for machine-readable output
  • Spec-driven Updates--spec FILE on create/update for batch metadata and section changes

Alfred is designed to be used by both AI agents (Claude Code, Codex, Gemini) and humans.

Quick Start

pip install fx-alfred
cd my-project
af guide          # see workflow routing (PKG → USR → PRJ)
af list           # list all documents
af read COR-1000  # read a specific document

Features

Workflow Routing (af guide)

Scans three layers for routing documents and outputs a complete workflow guide:

af guide --root /path/to/project
═══ PKG: COR-1103 Workflow Routing ═══
  Intent-based router: ALWAYS → PRIMARY ROUTE → OVERLAYS
  Golden rules from all COR SOPs

═══ USR: ALF-2207 Workflow Routing USR ═══
  Cross-project user preferences

═══ PRJ: FXA-2125 Workflow Routing PRJ ═══
  Project-specific decision tree

Workflow Checklists (af plan)

Generate step-by-step checklists from SOPs — optimized for LLM consumption:

af plan COR-1102 COR-1602 COR-1500            # phased checklist from named SOPs
af plan --human COR-1102                       # human-readable format
af plan --task "implement FXA-XXXX PRP"        # auto-compose SOPs from tags (COR-1202)
af plan --task "..." --todo --graph            # flat TODO + ASCII + Mermaid graph
af plan --task "..." --with-skills             # append matching skill docs
af plan --task "..." --graph --graph-format=ascii    # ASCII only (terminal)
af plan --task "..." --graph --graph-format=mermaid  # Mermaid only (GitHub / Obsidian)
af setup                                       # suggested prompts for agent config

Auto-compose matches the task description against Task tags: SOP metadata (deterministic, no LLM), includes any SOP with Always included: true as a baseline (e.g., COR-1103 routing), and topologically orders the result via Workflow input/Workflow output edges with layer+ASCII tiebreaks.

--todo flattens all phases into one continuously-numbered checklist with [SOP-ID] provenance, ⚠️ gate markers, and 🔁 loop-start / 🔁 back to N.M (max K) markers driven by Workflow loops: SOP metadata.

--graph emits both an ASCII box-and-arrow diagram (Unicode-width aware, terminal-friendly) and a fenced Mermaid block (pasteable into GitHub / Obsidian). Use --graph-format={ascii,mermaid,both} to pick one.

--graph-layout={nested,flat} (v1.7.0+, ASCII-only) chooses the DAG shape. The default nested layout draws each SOP as an outer phase-box containing inner step-boxes with connectors. Intra-SOP loops render as a 🔁 → N.M max K annotation line inside the phase box. Cross-SOP loops — declared via Workflow loops: [{from: N, to: "PREFIX-ACID.step", ...}] metadata — render as a right-side vertical track (◄───┐ ... ───┘ max N) that spans phase boundaries. flat restores the v1.6.x layout (one phase-box per SOP, steps listed inside, intra-SOP loops only) for downstream tooling pinned on the legacy shape.

See COR-1202 (Compose Session Plan) for the canonical usage procedure.

Agent Helpers and Skills

Run explicit local helper code only after opting in:

ALFRED_AGENT_TOOLS=1 af agent call collect_review_pack --arg root=. --json
ALFRED_AGENT_TOOLS=1 af agent run scripts/check_release.py --json

af agent call loads public functions from <project-root>/.alfred/agent_helpers.py first, then ~/.alfred/agent_helpers.py. The exact ALFRED_AGENT_TOOLS=1 gate is checked before importing helper code.

Discover reusable skill documents without executing code:

af skill list
af skill list --task "release fx-alfred to pypi" --json
af skill read skill-release-to-pypi

A skill is a REF or SOP document with explicit Tags: skill metadata. af plan --task "..." --with-skills appends matched skills to the plan; JSON output adds recommended_skills and uses schema version 3.

See FXA-2237 for the full usage reference.

Document Validation (af validate)

Enforces document health across all layers:

af validate --root /path/to/project

Checks:

  • H1 format (# TYP-ACID: Title)
  • Per-type required metadata fields (Applies to, Last updated, Last reviewed, Status)
  • Status values against allowed set per document type
  • Change History table structure
  • SOP required sections (What Is It?, Why, When to Use, When NOT to Use, Steps)
207 documents checked, 0 issues found.

Document Management

# Create
af create sop --prefix FXA --area 21 --title "My SOP"
af create prp --prefix FXA --area 21 --title "My Proposal"
af create sop --prefix FXA --area 21 --title "My SOP" --spec fields.yaml  # from spec file

# Read
af read COR-1000                    # by PREFIX-ACID
af read 1000                        # by ACID only

# Update
af update FXA-2107 --status "Active"
af update FXA-2107 --history "Done" --by "Claude"
af update FXA-2107 --title "New Title" -y
af update FXA-2107 --spec patch.yaml  # batch update via spec file

# Format
af fmt                              # show diff for all PRJ documents
af fmt FXA-2107                     # show diff for one document
af fmt --write                      # apply canonical formatting in-place
af fmt --check                      # CI check: exit 1 if any changes needed

# Search
af search "validation"              # search content across all docs
af search "validation" --json       # JSON output

# List & Filter
af list --type SOP                  # filter by type
af list --tag release               # filter by tag
af list --prefix FXA --json         # filter + JSON output

# Other
af guide --json                     # routing guide as JSON
af validate --json                  # validation results as JSON
af status                           # document counts by type/layer
af index                            # regenerate project index
af changelog                        # view version history

# Where (file path lookup)
af where FXA-2107                   # print absolute path
af where FXA-2107 --json            # JSON: {doc_id, path, source, filename}
vi $(af where FXA-2107)             # composable with shell tools

Three-Layer Document Model

graph TD
    AF["af CLI"] --> PKG
    AF --> USR
    AF --> PRJ

    subgraph PKG ["📦 PKG (read-only)"]
        direction LR
        P1["COR-* SOPs"]
        P2["Templates"]
        P3["Bundled with fx-alfred"]
    end

    subgraph USR ["👤 USR (~/.alfred/)"]
        direction LR
        U1["Personal preferences"]
        U2["Cross-project rules"]
        U3["Routing: ALF-2207"]
    end

    subgraph PRJ ["📁 PRJ (./rules/)"]
        direction LR
        R1["Project SOPs & CHGs"]
        R2["PRPs & ADRs"]
        R3["Routing: FXA-2125"]
    end

    style PKG fill:#1a1a2e,stroke:#e94560,color:#fff
    style USR fill:#1a1a2e,stroke:#0f3460,color:#fff
    style PRJ fill:#1a1a2e,stroke:#16213e,color:#fff
Layer Location Writable Scope
PKG Bundled in package No Universal COR documents
USR ~/.alfred/ Yes Personal, cross-project
PRJ ./rules/ Yes Project-specific

Document Types

Type Purpose Example
SOP Standard Operating Procedure How to create a document
PRP Proposal Design for a new feature
CHG Change Request Modify existing system
ADR Architecture Decision Record Record a decision
REF Reference Glossary, index, contract
PLN Plan Execution schedule
INC Incident Bug report, outage record

Document Format

<PREFIX>-<ACID>-<TYP>-<Title-With-Hyphens>.md

FXA-2134-PRP-AF-Plan-Command-Workflow-Checklist.md
COR-1103-SOP-Workflow-Routing.md

For AI Agents

Session Start

af guide --root /path/to/project    # 1. See routing + decision tree
af plan COR-1102 COR-1602 COR-1500 # 2. Generate workflow checklist

First Time Setup

af setup                            # See suggested prompts for your agent config

Decision Tree (COR-1103)

graph TD
    START["What are you doing?"] --> Q1{"Pure document\nmanagement?"}
    START --> Q2{"Something\nbroken?"}
    START --> Q3{"New capability\nthat doesn't exist?"}
    START --> Q4{"Change existing\nsystem?"}
    START --> Q5{"Record a\ndecision?"}
    START --> Q6{"Track/discuss\na topic?"}

    Q1 -->|New SOP| COR1000["COR-1000"]
    Q1 -->|New doc| COR1001["COR-1001"]
    Q1 -->|Update| COR1300["COR-1300"]

    Q2 -->|Bug| INC["INC"]
    Q2 -->|Fix + change| INC2["INC + CHG"]

    Q3 --> PRP["PRP (COR-1102)"]
    PRP --> REV["Review (COR-1602)"]
    REV --> CHG1["CHG (COR-1101)"]
    CHG1 --> TDD1["TDD (COR-1500)"]

    Q4 -->|Standard| CHG2["CHG, no review"]
    Q4 -->|Normal| CHG3["CHG → Review → TDD"]
    Q4 -->|Emergency| CHG4["CHG → Execute → Post-approval"]

    Q5 --> ADR["ADR (COR-1100)"]
    Q6 --> DT["COR-1201\nD new / D list"]

    style START fill:#0f3460,stroke:#e94560,color:#fff
    style PRP fill:#16213e,stroke:#0f3460,color:#fff
    style REV fill:#1a1a2e,stroke:#e94560,color:#fff
    style TDD1 fill:#16213e,stroke:#0f3460,color:#fff

Key SOPs

SOP What it does
COR-1103 Workflow routing — which SOP to follow for any task
COR-1202 Compose Session Plan — af plan --task … --todo --graph for full session workflow
COR-1102 Create Proposal (PRP lifecycle)
COR-1101 Submit Change Request (CHG)
COR-1500 TDD Development Workflow
COR-1602 Multi-Model Parallel Review
COR-1612 Respond to PR review comments on GitHub
COR-1615 GitHub App PR Review Bot Loop — trigger/poll/match-head loop for Codex Connector, Copilot, and other GitHub App reviewers
COR-1608 PRP Review Scoring rubric
COR-1611 Reviewer Calibration Guide
COR-1613 Council Review — multi-reviewer decision mechanism (14 voting rules)
COR-1503 Diagnose Feedback Loop — disciplined bug/perf diagnosis

Review Scoring

Alfred includes a standardized review scoring framework:

  • COR-1608 — PRP scoring (6 weighted dimensions + OQ hard gate)
  • COR-1609 — CHG scoring (5 dimensions)
  • COR-1610 — Code scoring (5 dimensions)
  • COR-1611 — Shared reviewer calibration guide

Pass threshold: >= 9.0/10. All deductions must cite specific lines.

Commands Reference

af guide [--root DIR] [--json]
af plan [SOP_ID ...] [--root DIR] [--task TEXT] [--with-skills] [--todo] [--graph] [--graph-format ascii|mermaid|both] [--human] [--json]
af agent call HELPER_NAME [--arg key=value ...] [--json]
af agent run SCRIPT_PATH [--json]
af skill list [--task TEXT] [--layer PKG|USR|PRJ|all] [--json]
af skill read ID_OR_NAME [--json]
af list [--type TYPE] [--prefix PREFIX] [--source SOURCE] [--tag TAG] [--json]
af read IDENTIFIER [--json]
af create TYPE --prefix P --acid N|--area N --title T [--layer project|user] [--subdir DIR] [--spec FILE] [--dry-run]
af update IDENTIFIER [--status STATUS] [--field KEY VALUE] [--history TEXT] [--by NAME] [--title TITLE] [-y|--yes] [--dry-run] [--spec FILE]
af fmt [DOC_IDS...] [--write] [--check]
af where IDENTIFIER [--json]
af search PATTERN [--json]
af validate [--root DIR] [--json]
af setup
af status [--json]
af index
af changelog

Install / Upgrade

pip install fx-alfred              # install
pipx install fx-alfred             # install (isolated)
pipx upgrade fx-alfred             # upgrade

Changelog

See CHANGELOG.md or run af changelog.

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

fx_alfred-1.10.0.tar.gz (960.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

fx_alfred-1.10.0-py3-none-any.whl (271.1 kB view details)

Uploaded Python 3

File details

Details for the file fx_alfred-1.10.0.tar.gz.

File metadata

  • Download URL: fx_alfred-1.10.0.tar.gz
  • Upload date:
  • Size: 960.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for fx_alfred-1.10.0.tar.gz
Algorithm Hash digest
SHA256 11ff425ee2f0bbca1f716120a7f7058e6df734dfa034af6aee246dc3a0af551a
MD5 3ec5fea509885850debc0d69cd9d150d
BLAKE2b-256 4619606ceb5d49a65b44eac74c458adcb9b465d766e8d0c0e627c21129010186

See more details on using hashes here.

Provenance

The following attestation bundles were made for fx_alfred-1.10.0.tar.gz:

Publisher: publish.yml on frankyxhl/alfred

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file fx_alfred-1.10.0-py3-none-any.whl.

File metadata

  • Download URL: fx_alfred-1.10.0-py3-none-any.whl
  • Upload date:
  • Size: 271.1 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for fx_alfred-1.10.0-py3-none-any.whl
Algorithm Hash digest
SHA256 9fb8d54ab34e672be31c00231d5cf5d53981eef9f4d4ce7cffa3a129d78c692a
MD5 e52d063e495db5060f6fdc0045a48ee0
BLAKE2b-256 9f01ae1058456f4f377717b6826d6a9530c402eca0865e6613f38e50c5159b90

See more details on using hashes here.

Provenance

The following attestation bundles were made for fx_alfred-1.10.0-py3-none-any.whl:

Publisher: publish.yml on frankyxhl/alfred

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page