Implementation BlueprintExecution layer

wcagent Verifiable Code Delivery Retainer (7-10 days)

Agency retainer that deploys wcagent inside client VS Code environments so every AI-assisted code change ships with test, build, and diff evidence the client can audit. Priced as a per-developer monthly service on top of the $4.99 USD seat cost. Time: 7-10 days.

By InnovaAI ResearchPublished Updated

AI Code Tools

How do you implement it?

Blueprint

wcagent Verifiable Code Delivery Retainer (7-10 days)

Agency retainer that deploys wcagent inside client VS Code environments so every AI-assisted code change ships with test, build, and diff evidence the client can audit. Priced as a per-developer monthly service on top of the $4.99 USD seat cost.

Prerequisites
  • Client development team works in desktop VS Code, since wcagent is a VS Code extension and has no other editor surface
  • At least one eligible AI provider account (ChatGPT, Claude, Gemini, Grok, or DeepSeek) already funded by the agency or client, because wcagent does not bundle model access
  • Client repository has a runnable test suite or build command, otherwise completion gating has nothing to verify against
  • Written client sign-off on local execution approval rules covering commands, file edits, and tool calls
  • A named client-side approver who can review and release gated actions during the engagement window
Execution Timeline
  • 1.Install the wcagent extension in the client's desktop VS Code and confirm the seat is active at $4.99 USD per month
  • 2.Connect one eligible provider account (ChatGPT, Claude, Gemini, Grok, or DeepSeek) through an authorized method and verify the connection returns a response
  • 3.Record which provider the client is paying for separately so the retainer invoice does not imply bundled model access
  • 1.Configure local developer approval settings so commands, file edits, and tool execution are gated before any client repository is touched
  • 2.Walk the client approver through the approval prompt flow and agree on which action classes auto-approve versus require a human
  • 3.Document the approval matrix as a one-page artifact the client keeps after handoff
  • 1.Point wcagent at the client repository and author an AGENTS.md file with project-level instructions
  • 2.Tune repository context selection so the context engine ranks only task-relevant files, symbols, and diagnostics
  • 3.Run a first low-risk task and inspect which files the context engine selected to confirm ranking behaves as expected
  • 1.Wire the client's existing test command and build command into the completion gate
  • 2.Run a deliberately failing task to confirm wcagent blocks completion when tests fail rather than accepting model confidence
  • 3.Capture the blocked-run evidence as a demo the client can replay
  • 1.Execute three real backlog tickets through wcagent with full evidence capture
  • 2.Review the recorded diffs and diagnostics with the client approver and log any rejected changes
  • 3.Adjust AGENTS.md instructions based on the rejection patterns from day five
  • 1.Test task coordination and recovery by interrupting a run mid-task and confirming wcagent resumes without losing context
  • 2.Verify the complete evidence history is queryable for each of the three tickets
  • 3.Identify any repository areas where context ranking underperforms and note them for the handoff summary
  • 1.Build the client-facing evidence report template that maps each merged change to its test result, build status, and diff
  • 2.Train two client developers on the approval workflow and the evidence history view
  • 3.Agree the monthly cadence for evidence review and the escalation path when a change is blocked
  • 1.Run a supervised shift where the client developer drives wcagent and the agency observes
  • 2.Fix any approval friction surfaced during the supervised shift
  • 3.Finalize the handoff summary covering configuration, provider connections, and approval rules
  • 1.Confirm the client can independently install wcagent on a second developer machine using the documented configuration
  • 2.Reconcile the seat count against the client's developer headcount so billing matches actual usage
  • 3.Close out with a signed acceptance that the evidence history meets the client's audit requirement
$1,800 to $4,500 USD for the 7-10 day deployment, plus $4.99 USD per developer per month for wcagent seats and the client's separate AI provider subscription cost7-10 days
ROI Logic

The wcagent seat costs $4.99 USD per developer per month, so a five-developer client generates under $25 USD in tool cost against a retainer that agencies can price from $1,800 USD for the initial deployment and a monthly fee thereafter. Margin comes from the configuration and evidence-reporting labor, not the license, which means the agency keeps the majority of every retainer dollar while the client's only hard tool cost is the seat plus their own AI provider bill. The constraint is that clients must already run VS Code with working tests and builds, so the offer sells best into teams that have that infrastructure and worst into greenfield shops.

Deliverables
  • Configured wcagent installation across the client's developer machines with provider connections verified
  • AGENTS.md project instruction file tuned to the client repository
  • Approval matrix documenting which commands, edits, and tool executions require human release
  • Evidence report template mapping each merged change to its test result, build status, and diff
  • Handoff summary covering configuration, seat count, and the monthly evidence review cadence
Definition of Done

The client's own developers can run wcagent tasks end to end, and every completed change in the evidence history carries a passing test result, a successful build, and a recorded diff without agency intervention.

Open the wcagent coursewcagent decision recordBrowse the Academy

More on wcagent

More for AI Code Tools