WhatHaveIDone (WHID)

    1

    Understand code changes, trace cross-repo impact, and learn unfamiliar projects with repository evidence.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    WhatHaveIDone (WHID)

    Example session with this skill installed

    Use WHID Quick to analyze commit a307056 in hasan-gundogdu/WhatHaveIDone. Explain what changed, why it matters, and what remains unverified. Do not modify files or run tests.

    • Read your context and instructions
    • Compiled the whathaveidone whid

    WhatHaveIDone — Quick

    WHID Core version: 0.2.0-alpha.2

    Repository: hasan-gundogdu/WhatHaveIDone
    Scope: a67f309 → a307056 (target commit compared with its first parent).

    What changed

    You made the existing WHID skill distributable as three marketplace packages. The core analysis instructions did not change: the diff contains no paths under skills/whid/. CONFIRMED — endpoint diff a67f309..a307056.

    Critical changes

    One source, three upload formats. The new scripts/build-marketplace-packages.ps1 builds Codex and Claude plugin ZIPs plus an Agensi skill ZIP. It takes the canonical SKILL.md and references from skills/whid/; Agensi receives them at the archive root. DIRECT impact: each platform receives its expected package layout. CONFIRMED — builder entry allowlist and target loop at a307056.
    2.

    Publication metadata is separate from behavior. Root plugin.json describes the portable/OpenAI package; .claude-plugin/plugin.json identifies the Claude plugin. Both carry version 0.2.0-alpha.2. The Claude marketplace entry points to the repository root. DIRECT impact: plugin discovery and packaging, while the shared skill stays vendor-neutral. CONFIRMED — those three manifests at a307056.
    3.

    The builder checks the archive, not just the inputs. It checks version agreement, invokes repository validation, then reads every ZIP entry back and compares its bytes with the allowlisted source. It writes hashes and build provenance. DIRECT impact: missing or altered packaged content can stop the build. CONFIRMED — archive readback loop and build-info.json generation at a307056.
    4.

    Data handling is described explicitly. PRIVACY.md distinguishes WHID instructions from the host agent’s data policy and explains saved reports. CONFIRMED — PRIVACY.md at a307056.

    Limits and what to remember

    UNKNOWN: whether a marketplace will approve or publish the package. Manifest files and archive checks do not prove review acceptance, semantic report quality, or cross-agent compatibility. A matching version identifies the contract; it does not certify its behavior.

    Tests were not run for this analysis. This example explains a real public commit; it does not claim a completed cross-agent release evaluation.

    ZIPs plus an Agensi skill ZIP. It takes the canonical SKILL.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    AI wrote the code. Make sure you understand it.

    WhatHaveIDone (WHID) helps developers recover ownership of repository changes and learn unfamiliar projects. It explains behavior, execution flow, architecture, and impact using repository evidence. Free and open source under Apache-2.0.

    Three ways to understand your code

    • Quick: a focused ownership recap of your working-tree changes, one commit, or a revision range.
    • Deep: Before → After behavior, execution flow, critical code, architecture reasoning, test evidence, contextual concepts, and understanding questions.
    • Onboard: an ordered learning route through one current project. Optionally save a Markdown guide, then refresh affected chapters or resume unfinished work.

    Single repo or explicit multi-repo

    Use --repo for one repository, --repos api,web for selected repositories, or --all for the authorized workspace. Multi-repo Quick/Deep examines each working tree separately, then connects provider contracts and consumer expectations only where evidence supports the connection. Multi-repo commit/range analysis and multi-repo onboarding are not supported.

    Evidence and control

    Findings distinguish CONFIRMED, INFERRED, and UNKNOWN. Quick/Deep are read-only. Saved Onboard writes only managed guide files in the explicitly requested directory. Tests run only if separately requested. Reports follow your language. WHID is explicitly invoked and does not run after ordinary coding tasks on its own.

    Requirements and limits

    Requires Git and a capable coding-agent environment with repository access. Codex and Claude Code are the primary validation targets; other compatible Agent Skills clients are best effort. This is Core 0.2.0-alpha.2, a prerelease; completed cross-agent release certification is not claimed. The host agent provides repository inspection and file operations. There is no WHID-operated backend or separate WHID API key. Your host agent’s usage charges and data policies still apply.

    Ignored files, binaries, generated output, and credential candidates are excluded unless separately authorized. Current instructions govern safe operation; historical analysis uses revision-correct repository evidence.

    Source, full usage, and support: https://github.com/hasan-gundogdu/WhatHaveIDone Privacy: https://github.com/hasan-gundogdu/WhatHaveIDone/blob/main/PRIVACY.md

    Core 0.2.0-alpha.2 is a prerelease. Cross-agent release certification is pending. Git and repository inspection are required. Multi-repo supports Quick/Deep working trees only, not commit/range scopes or onboarding. A...B scopes are rejected. Ignored files, binaries, generated output, and credential candidates are excluded unless separately authorized. The host model can still make mistakes; evidence labels are not a guarantee. Saved Onboard additionally needs scoped file writes and SHA-256 hashing.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 27 days ago

    • Free with an Agensi account

    Requires

    Git

    Permissions

    Read Files
    Write Files
    Terminal / Shell

    Read selected Git repositories, revision-local code/configuration/documentation, and relevant tests. Terminal/Git inspection is read-only; tests or setup run only on a separate user request. Write access is optional: explicitly requested saved reports only (Quick/Deep outside analyzed repos; Onboard in its dedicated requested directory). No WHID network service or environment-variable access is required. The host model provider may process inspected content under its own data policy.

    Listed27 days ago

    What's inside

    Frequently Asked Questions

    It explains what changed at the behavior level, how the changed code participates in execution flow, what else it may affect, and the concepts needed to understand it. Quick is concise; Deep adds a walkthrough, evidence appendix, and contextual questions.

    A coding-agent environment with Git and repository inspection access. Saved Onboard additionally needs scoped file writes and SHA-256 hashing. WHID supplies instructions and references; it does not add Git or repository access to a chat that lacks them.

    Codex and Claude Code are the primary validation targets. Other clients that support Agent Skills and repository inspection are best effort. This prerelease does not claim completed cross-agent certification. Your existing agent provider supplies the model.

    No. WHID instructs the agent to distinguish CONFIRMED, INFERRED, and UNKNOWN claims and cite repository evidence. This makes claims easier to verify but does not prevent all model mistakes. Review important conclusions against the cited code and configuration.

    Yes. Explicit --repos or --all selection supports Quick/Deep working-tree analysis across authorized workspace roots. Each Git history remains independent. Cross-repo findings need actual contract, caller, configuration, or integration evidence. Multi-repo historical scopes and onboarding are not supported.

    The skill is free under Apache-2.0; your coding-agent usage costs still apply. Quick/Deep do not modify analyzed repositories or Git state. Explicitly saved reports can write only to permitted report destinations; saved Onboard can manage guide files in its requested directory. Tests need a separate request.

    New in Technical Documentation