MissionsIllustrative mission, phase by phase

A multi-repo mission, from objective to review.

An illustrative run through all six phases: discovery, plan approval, delegation, isolated implementation, independent review, and what the mission keeps afterwards.

One mission can coordinate AI coding agents across multiple repositories. Each specialist works in an isolated worktree, with terminal, browser, and automations attached to the same objective.

Mission manifest

rl-311 · per-tenant rate limiting

illustrative example

Objective

Introduce per-tenant rate limiting across the public API surface.

──no breaking change for existing clients

──limits configurable per plan tier

initiated
example operator
registered checkouts
4
agents
4 + verifier
approvals
1 gate · infra
Acceptedexample state
illustrative topology4 isolated worktrees·read-only review·contract checks included

Illustrative product walkthrough, not measured customer evidence

01Discoverphase 01

Read the workspace before touching it.

The mission opens read-only. It walks the registered checkouts, works out which of them the objective actually reaches, and pulls any memory from earlier missions that applies.

mode
read-only · no changes yet
impact map
observed before planning
found
registered checkouts · relevant memory
systems map / scan2/6 in scope
api-gatewaycheckout

enforcement point

billing-servicecheckout

plan tiers · usage

········scanning

usage dashboard

········scanning

redis counters

········scanning

build + deploy matrix

········scanning

client retry behavior

reading imports, configs, and pipeline definitions…
02Planphase 02

One plan across four repositories.

The mission drafts a single plan covering every registered checkout, pins the interface all four will depend on, and blocks in plan mode until you approve the boundary.

workstreams
4 · one per repository
contract
rate-limit headers · pinned v1
gate
plan approval → human decision
plan / rl-3114 workstreams

Pinned interface · v1

RateLimit-Limit · RateLimit-Remaining · RateLimit-Reset · 429 + Retry-After on exceed · limits.yaml per plan tier

w1api-gatewaytoken-bucket middleware · limit headers
w2billing-serviceplan-tier limits API · usage events
w3platform-infraredis counters · alerts · dashboards
w4web-consoleusage meter · limit banner

approval requested · new redis cluster · est. $84/mo

plan proposed

example approval · scope pinned · execution unlocked

plan approved
03Delegatephase 03

One specialist per repository.

Each specialist receives a packet naming its checkout, its slice of the objective, and the pinned contract, then works in its own git worktree. The parent mission keeps the objective, and a mutation-disabled reviewer is held back for the next phase.

agents
4 specialists + 1 reviewer
reviewer
file.read · file.search · agent.report
isolation
one repo per agent · shared contract

Lead orchestrator

Holds the objective and the pinned contract

api-gateway

Repo agent 1

middleware + headers

model = "frontier"

billing-service

Repo agent 2

tier limits + events

model = "frontier"

platform-infra

Infra agent

redis · alerts · IAM

model = "fast"

web-console

Repo agent 3

usage UI

model = "fast"

Reviewer · independent · mutation disabled
04Implementphase 04

Four worktrees that cannot collide.

Specialist changes stay in separate Git worktrees. The parent mission owns the seam between them, so integration is one decision rather than four repositories discovering each other late.

changes
4 bounded worktrees
integration
parent mission owns the seam
execution
linked Mac · registered repositories
isolated worktrees4 bounded · 1 objective
api-gatewaywt-01isolated specialist worktreebounded
billing-servicewt-02isolated specialist worktreebounded
platform-toolswt-03isolated specialist worktreebounded
web-consolewt-04isolated specialist worktreebounded
specialists remain isolated · integration stays in the parent missionexecution local · review remains separate
05Reviewphase 05

A reviewer with no write access.

The reviewer holds file.read, file.search, and agent.report, and runs with mutationAllowed set to false. It can read every worktree and report what it finds; it cannot change any of it.

reviewer
independent · mutation disabled
tools
file.read · file.search · agent.report
profile
read-only capability operations
evidence reviewRead-only
filessearchreviewreport

file.read

reviewer can inspect the implementation

file.search

reviewer can trace relevant references

agent.report

findings return to the parent mission

mutationAllowed: false

reviewer cannot change the files

read-only capability operations

the role boundary is explicit

reviewed by an agent that cannot edit the code · findings attached to the mission
06Rememberphase 06

Three records, each with its source.

As the mission completes, Skyflo drafts what it taught. Records backed by the work's own evidence are kept; anything weaker waits for you. Each one stays linked to the mission and files that produced it, and a workflow this mission verified can become a Skill once another repository verifies it too.

records
scoped · cited · revertible
recall
retrieved when relevant
compounds
proven twice · becomes a Skill
How Skyflo learns between missions
knowledge / commit3 records · evidence linked
Outcomemission rl-311

Public API rate limits enforced at api-gateway via token bucket in Redis. Limits per plan tier live in platform-infra/limits.yaml.

Constraintmission rl-311

429 responses must include Retry-After. SDK v2 clients retry automatically; v1 clients surface the error.

Ownershipmission rl-311

platform-infra owns the Redis cluster. billing-service owns plan-tier limit definitions.

recall“how do I raise limits for an enterprise tenant?”

→ retrieved from learned mission memory in a later session.

example complete·six product phases shown·one objective, four repositories, one review report

Other missions that span systems.

Cross-repository migrations, coordinated upgrades, security responses, and domain splits all need one approved objective to land consistently across several systems.

m.01migration

Move checkout workloads to ARM without interrupting traffic

6 systems · one approved plan

m.02security

Introduce tenant-level encryption across the platform

4 systems · one approved plan

m.03upgrade

Upgrade the framework major version in every affected service

9 systems · one approved plan

m.04response

Patch a critical CVE everywhere it is actually reachable

12 systems · one approved plan

m.05refactor

Split the monolith's billing domain into its own service

5 systems · one approved plan

m.06observability

Instrument p95 tracing across the request path

7 systems · one approved plan

How missions behave.

Scope, isolation, and review. For what a mission keeps once it closes, see how Skyflo learns between missions.

01What is a mission?+

A mission is the persistent local unit that holds one objective, its plan, specialist work, execution history, and useful context.

02Can one mission span repositories?+

A workspace is an ordered set of registered Git checkouts; one mission may span them.

03Can other coding agents work across repositories?+

Yes. Several coding agents and orchestration tools can work across repositories. Skyflo's distinction is the mission record that keeps the approved plan, bounded specialists, execution evidence, independent review, and accepted memory together.

04Does Skyflo publish pull requests?+

No current product claim promises PR publication. Specialists work in isolated Git worktrees and integrate through the local mission.

05How is the work reviewed?+

A separate reviewer can inspect files and report findings, but cannot edit files or run mutating capability operations.

Run one of these on your own repositories.

Skyflo Desktop is free with a signed-in coding agent or your own provider keys. Register the checkouts a change touches and give the mission the objective.

Missions run locally with the models you configure.