Skip to main content

agile-v-core

Apply Agentic Agile-V process control, task briefs, risk-based evidence gates, and human approval rules.

Jump to install

Source facts

Repository
Agile-V/agentic_agile_v
Last source activity
July 8, 2026 at 18:43
Detected SKILL.md language
English
Stars
7
Forks
3

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
agile-v-core
description
Apply Agentic Agile-V process control, task briefs, risk-based evidence gates, and human approval rules.
# Agile-V Core Rules ## Operating Principle Conversation is allowed for discovery. Implementation requires a structured task brief. Do not implement from an ambiguous chat history. Do not expand scope beyond the task brief. ## Before Editing Code 1. Locate the task brief (`.agentic-agile-v/tasks/AAV-XXXX/task_brief.md`) 2. Confirm risk level (L0-L4) 3. Confirm allowed and blocked paths 4. Confirm acceptance criteria 5. Produce a short plan and wait for acknowledgement on L2+ tasks ## Context Engineering Keep context lean. Pass file paths, not file contents, to sub-agents. Spawn a fresh context per major phase (requirements, build, verify). Size tasks to fit ≤50% of available context. When context exceeds 70%, stop and summarize state before continuing. ## Completion Criteria A change is complete only when the evidence bundle satisfies the required risk level. Generating code is not completion. Passing evidence validation is completion. ## Never - Remove tests to make a build pass - Weaken security controls (auth, crypto, input validation) - Add dependencies without approval - Modify public APIs unless explicitly requested - Self-approve high-risk work (L3/L4) - Expand scope beyond allowed paths — if you notice adjacent issues, log them as observations, do not fix them - Commit secrets, tokens, credentials, or personal data - Bypass task brief requirements - Claim tests pass without running them - Claim CI passes without evidence from CI output ## Evidence Requirements by Risk Level | Level | Tests | Verifier | Approval | Special | |-------|-------|----------|----------|---------| | L0 | Optional | No | No | - | | L1 | Required or rationale | No | No | - | | L2 | Passing tests | Yes | Reviewer | - | | L3 | Passing tests | Yes | Domain owner | Rollback path | | L4 | Passing tests | Yes | Formal approval | Traceability + simulation/HIL/formal | ## Scope Control Changed files must be within `allowed_paths` from the task brief. Files in `blocked_paths` must not be modified. Dependency changes require explicit approval in the task brief. If you identify a change needed outside `allowed_paths`, stop, document it as an observation in the evidence bundle, and ask the human whether to expand scope. ## Halt Conditions Stop and ask before proceeding when: - The task brief is missing or ambiguous - Acceptance criteria cannot be mapped to a concrete test - A required change is outside `allowed_paths` - Risk level is unclear (default to the higher level) - An L3/L4 change lacks explicit human approval ## Handoff On session end, generate a handoff summary: - Current objective - Changed files - Tests run and their results - Open risks - Next recommended action
View on GitHub