AI Assist Slice¶
Purpose¶
This page captures the first practical AI integration slice for gig.
The goal is not to turn gig into a generic chat CLI.
The goal is to let gig keep owning deterministic release evidence while an AI layer turns that evidence into a more executive, more readable briefing.
Core Rule¶
gig remains the source of truth.
That means:
gigcomputes ticket scope, branch comparisons, risk signals, dependency status, and verification verdicts- AI consumes the resulting bundle
- AI explains and prioritizes
- AI does not invent missing branch or commit facts
Why DeerFlow Fits¶
DeerFlow already has the pieces that are useful on top of gig:
- a lead agent with tool orchestration
- optional sub-agents for deeper follow-up work
- memory and skill support for reusable workflows
- a simple HTTP surface for thread creation and streaming responses
That makes it a good sidecar for:
- executive summaries
- release-manager briefings
- evidence-gap callouts
- recommended next commands
- future multi-source evidence synthesis across SCM, CI, deployments, and issue trackers
Boundary¶
The integration boundary should stay clean:
internal/inspect,internal/plan, andinternal/manifestkeep their current deterministic responsibilitiesinternal/assistantbuilds an audit bundle from those services- the DeerFlow client receives that bundle and asks for a briefing
- CLI output renders the final AI summary, but the structured
gigresult still exists underneath
This keeps the CLI thin and avoids pushing model-specific logic into the release engine.
Current Slices¶
The current shipped commands are:
gig assist doctor
gig assist setup
gig assist audit --ticket ABC-123 --repo github:owner/name --audience qa
gig assist release --release rel-2026-04-09 --path . --audience release-manager
gig assist release --release rel-2026-04-09 --ticket-file tickets.txt --repo github:owner/name --audience release-manager
gig assist resolve --path . --ticket ABC-123 --audience release-manager
gig assist doctor:
- finds the bundled
deer-flow/workspace inside thegigrepo - checks whether the local sidecar is configured, startable, and reachable
- reports the next onboarding step without writing files
gig assist setup:
- finds the bundled
deer-flow/workspace inside thegigrepo - creates
config.yamland any available DeerFlow env templates when needed - tells the user the recommended next start command for the local sidecar
gig assist audit:
- resolves repo scope and branch context using normal
gigrules - builds a deterministic bundle from inspection, planning, verification, and manifest highlights
- sends that bundle to DeerFlow
- prints a concise AI briefing tuned for
qa,client, orrelease-manager
gig assist release:
- loads saved ticket snapshots from one named release, or captures a live ticket set from local or remote repositories
- builds a deterministic release bundle with release-plan rollups and packet data
- sends that bundle to DeerFlow
- prints a concise AI release briefing for the selected audience
gig assist resolve:
- loads the deterministic
gig resolvestatus plus the first active supported conflict block - builds a conflict bundle with provenance, risk, scope warnings, and supported resolver actions
- sends that bundle to DeerFlow
- prints a concise AI conflict briefing for the selected audience
These commands are intentionally experimental. It is an additive explanation layer, not a required step in the main product flow.
What This Slice Solves¶
This slice helps with the practical pain points that deterministic JSON alone does not solve well:
- stakeholders want a short summary, not raw repo data
- release managers want obvious next commands
- teams want risk phrased in business-readable language
- people want the same facts explained differently for QA, client review, and release handoff
- release owners want one release-wide summary across many ticket snapshots or live ticket sets instead of reading each ticket separately
- developers want help understanding which conflict choice is safest without replacing the deterministic resolver
What This Slice Does Not Do¶
It does not:
- replace
verify - replace
manifest generate - guess release state without
gigevidence - require DeerFlow for the main remote audit workflow
- change
giginto a write-enabled automation tool - replace
gig resolve startas the actual place where conflict choices are applied
Skill Support¶
The repo now includes project-specific skill docs so DeerFlow or other agents can stay aligned with gig:
- Agent Skills
deer-flow/skills/custom/gig-release-audit/SKILL.mddeer-flow/skills/custom/gig-resolve-conflict/SKILL.mddeer-flow/skills/custom/gig-product-guardrails/SKILL.md
These skills reinforce the main boundary:
- use
gigas the evidence engine - use AI to explain and prioritize
- do not let prompts replace branch, ticket, or risk reasoning
- support both saved-snapshot and live remote release bundle workflows
Near-Term Direction¶
After the current slices, the next high-value steps are:
- expand the release-level bundle with richer cross-ticket evidence such as PR, CI, deployment, and issue-tracker context
- make repeated AI briefs work more naturally through saved workareas so users need fewer scope flags
- expose
gigitself as a tool or MCP surface so DeerFlow can request fresh evidence instead of relying only on prompt text - add richer prompt templates inside the project skill pack without moving logic into prompts
Product Positioning¶
The product story becomes stronger with this framing:
gig is a deterministic release-audit engine with an optional AI briefing layer.
That is more professional than a generic AI assistant and more useful than a plain branch-comparison tool.