Knowledge Contour Skills

Language: English | Русский

This repository contains ready-to-install Codex skills: self-contained folders with SKILL.md, agent prompt files, references, scripts, and tests.

Skills

product-workflow

End-to-end product shaping: discovery, decision log, PRD, user scenarios, current-scenario baseline, feature impact analysis, implementation plan, hardening plan, independent verification, editorial pass, and stakeholder-facing PDF.

Use it for product or feature descriptions, roadmap validation, scenario design, and implementation planning. For existing products, the baseline mode captures current scenarios as current-scenario-baseline.md, scenario-cards.md, and scenario-graph.dot; new increments then declare affected scenarios through pre-scan and impact artifacts before implementation planning. By default, the PDF includes only the product problem, scenarios, chosen solution, and independent verdict; T/H plans remain engineering artifacts.

flowchart TD
    U["User vision"] --> D["Phase 0: discovery agents"]
    D --> DL["Decision log: proposed / approved / delegated"]
    DL --> H{"Human approval?"}
    H -->|approved| S["Scenarios + overview"]
    H -->|needs choice| U
    S --> B["Baseline: scenario cards + DOT graph"]
    B --> I["Increment pre-scan + impact"]
    I --> C["Scenario critic loop"]
    C --> P["Implementation plan with Product artifacts"]
    P --> HP["Hardening plan"]
    HP --> V["Independent artifact verifier"]
    V -->|blockers| C
    V -->|approved verdict| E["Editorial style pass"]
    E --> PDF["Product PDF"]
    P --> IV["Post-implementation verifier"]
    HP --> IV

Loading

Contents:

• SKILL.md — main workflow and gates.
• agents/ — discovery, critic, verifier, and style-editor prompts.
• references/ — templates for scenarios, baseline, scenario cards, increment pre-scan/impact, decision log, plans, and PDF.
• scripts/verify_artifacts.py — structural checks for scenarios, baseline/pre-scan/impact artifacts, plans, hardening plans, and validation gate.
• scripts/build_pdf.sh — PDF assembly with mandatory independent validation.
• evals/ — expected-behavior eval set.

service-knowledge-contour

Minimal knowledge contour for one service repository: startup docs, canonical SERVICE_MAP.md / VERIFY.md, knowledge-gap registry, generated overlays, audit, promotion, and pruning.

Use it when a service needs a stable operating knowledge layer for humans and agents, onboarding docs are missing or fragmented, or topology, entrypoints, verification commands, integrations, or risk zones changed.

flowchart TD
    R["Repository reality"] --> B["bootstrap"]
    B --> C["Canonical core"]
    C --> G["Generated layer"]
    G --> A["audit / strict audit"]
    A -->|trigger fired| H{"Human approval needed?"}
    H -->|yes| P["promote / repair candidate"]
    H -->|no| G
    P --> C
    C --> V["Independent contour verifier"]
    A --> PR["PR / CI evidence"]

Loading

Contents:

• SKILL.md — service knowledge contour workflow and rules.
• agents/contour-verifier.md — independent semantic contour verification.
• bin/ — bootstrap, refresh, audit, promote, and prune shell scripts.
• examples/ — GitHub Actions and PR template examples.
• tests/ — bootstrap/audit contract tests.

Install In Codex

Install a skill by pointing Codex at the repository folder URL.

Ask Codex:

Install the skill from https://github.com/ehlyzov/skills/tree/main/product-workflow

or:

Install the skill from https://github.com/ehlyzov/skills/tree/main/service-knowledge-contour

Manual install:

mkdir -p ~/.codex/skills
cp -R product-workflow ~/.codex/skills/
cp -R service-knowledge-contour ~/.codex/skills/

Update an installed copy:

rm -rf ~/.codex/skills/product-workflow ~/.codex/skills/service-knowledge-contour
cp -R product-workflow service-knowledge-contour ~/.codex/skills/

Verify installation:

test -f ~/.codex/skills/product-workflow/SKILL.md
test -f ~/.codex/skills/service-knowledge-contour/SKILL.md

Install In Claude Code

Add this repository as a Claude Code plugin marketplace:

/plugin marketplace add ehlyzov/skills

or, with the full Git URL:

/plugin marketplace add https://github.com/ehlyzov/skills.git

Then install one or both plugins from the marketplace:

/plugin install product-workflow@knowledge-contour-skills
/plugin install service-knowledge-contour@knowledge-contour-skills

The marketplace file is stored at .claude-plugin/marketplace.json, which is the path Claude Code expects when adding a GitHub repository as a plugin marketplace.

Run Scripts

Running scripts is separate from installing a skill.

For product-workflow, run scripts from the skill folder or by absolute path to the installed skill:

python3 product-workflow/scripts/verify_artifacts.py --phase scenarios <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase baseline <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase pre-scan <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase impact <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase plan <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase hardening <repo-root>
python3 product-workflow/scripts/verify_artifacts.py --phase validation <repo-root>
bash product-workflow/scripts/build_pdf.sh <repo-root> ~/Downloads/product-docs.pdf

--phase pre-scan and --phase impact are explicit increment checks: they require corresponding files under docs/product/increments/. --phase all accepts a baseline-only product snapshot with no active increment files.

New T/H plans must include Product artifacts in every task. Existing legacy plans can be checked during migration with:

python3 product-workflow/scripts/verify_artifacts.py --phase all <repo-root> --allow-legacy-plan

build_pdf.sh requires a fresh docs/product/validation/verdict.md by default and excludes implementation/hardening plans. Use an explicit flag for internal engineering PDFs:

INCLUDE_ENGINEERING_PLANS=1 bash product-workflow/scripts/build_pdf.sh <repo-root> ~/Downloads/internal-product-docs.pdf

For service-knowledge-contour, scripts are copied or run inside the target service repository as an operating toolchain:

./bin/bootstrap.sh
./bin/refresh_contour.sh --check
./bin/audit_contour.sh --strict
./bin/promote_learning.sh --input-file /tmp/learning.txt
./bin/prune_contour.sh

Do not copy the whole skill folder into a service repository. Target service repositories should receive only the needed bin/* scripts or the contour created by bootstrap, not SKILL.md, tests, and prompt files.

Verify This Repository

python3 -m py_compile product-workflow/scripts/verify_artifacts.py
bash -n product-workflow/scripts/build_pdf.sh service-knowledge-contour/bin/*.sh
pytest -q tests/product_workflow service-knowledge-contour/tests

Basic skill-folder validation:

python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py product-workflow
python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py service-knowledge-contour

If PyYAML is missing in the current environment, install it into the active Python env or run validation in the Codex environment where dependencies are available.

Maintenance Rules

• The canonical entry point for every skill is SKILL.md.
• All agent prompt files in agents/ are written in English.
• If the skill request or target artifacts are Russian-language, agents must produce polished Russian user-facing text.
• Additional materials should live in agents/, references/, scripts/, assets/, examples/, tests/, or evals/.
• Do not add README files inside skill folders without a separate reason; this root file documents the skill set.
• Scripts must stay executable when the workflow invokes them directly.
• Generated overlays and PDFs are not source of truth and must not replace the markdown canon.