Processkit

Standalone MCP/harness package for:

• /business-process-trace — brownfield cross-system process evidence
• /business-impact-review — vertical × horizontal code-change review
• /configure-repo-maps — NL → machine-local checkout maps (cross-repo)
• deprecated /flow-trace redirect

Optional accelerators: CodeGraph, Docskit and ArtifactGraph. Processkit works without them through targeted repository search/model analysis. Missing optionals produce one schema-validated evidence event per run/optional with actual file-read and context-byte metrics; see docs/OPTIONAL-ACCELERATORS.md.

Quick start (member)

curl -fsSL https://raw.githubusercontent.com/raintr91/Processkit/main/install.sh | bash
cd /path/to/project
processkit init        # wizard: agents → lane (docs|fe|be)

The wizard picks agents (checkbox, detected ones pre-checked), then the lane. Every selected agent gets a project-local MCP config in the current repo; there is no location prompt. Init also:

• ensures platform-repos.local.json and legacy-repos.local.json when missing (never overwrites; never seeds portable *-repos.json);
• merges generated local targets into .gitignore (idempotent; unanchored patterns .cursor/ / .processkit/ matching Platform DNA / Codegenkit);
• installs shared cross-repo-index.mdc (DNA SSOT filename — no duplicate processkit-cross-repo-index.mdc);
• records entries in the install manifest so status can report missing ignore lines and empty local maps.

CI keeps the long flags:

processkit init --type=docs --target=cursor --yes

Local checkout / global CLI

From a local checkout:

pnpm install && pnpm build
# one-shot init without global install:
node /path/to/Processkit/bin/processkit.mjs init

# or refresh ~/.processkit + PATH shims from this checkout:
./install.sh --from /path/to/Processkit
# equivalent: PROCESSKIT_SRC=/path/to/Processkit ./install.sh

When Platform DNA should invoke this checkout instead of an older global processkit on PATH:

export PLATFORM_DNA_PROCESSKIT_ROOT=/path/to/Processkit
platform-dna init

Profiles:

• docs: process trace + impact review + configure-repo-maps + deprecated redirect
• fe / be: impact review + configure-repo-maps

Cross-repo routing

Step system id	Checkout map
legacy-*	legacy-repos.local.json
otherwise	platform-repos.local.json

In-repo-only work needs no map. Cross-repo with empty/missing keys → Gaps + /configure-repo-maps, then platform-dna codegraph:wire.

Example prompts (/configure-repo-maps)

Platform multi-repo (writes platform-repos.local.json):

/configure-repo-maps
2 portal: admin at ~/ws/portal-admin, line at ~/ws/portal-line;
2 API: core at ~/ws/api-core, scenario at ~/ws/api-scenario;
docs = ~/ws/base-docs; tests = ~/ws/base-tests.

Legacy (writes legacy-repos.local.json):

/configure-repo-maps
Legacy ERP is at D:\legacy\erp, key legacy-erp.

Then run /business-process-trace or /business-impact-review as usual.

Managed harness lifecycle

init records only Processkit-managed harness assets and their installed hashes in .processkit/install-manifest.json. Switching profiles marks assets from the previous profile as stale; it does not delete them or manage shared registries and portable project maps. It also records the destination in the XDG install ledger at $XDG_STATE_HOME/processkit/installs.json (or ~/.local/state/processkit/installs.json).

processkit status --project-root /path/to/project
processkit prune --project-root /path/to/project        # dry-run
processkit prune --project-root /path/to/project --yes  # delete safe stale files
processkit deinit                                       # current repo + local MCP
processkit uninstall                                    # every repo + MCP + CLI

Prune deletes only stale files still matching their recorded hash. Customized files are retained. platform-repos.json, shared extract registries and any file absent from the Processkit install manifest are never prune targets. Unsupported manifest APIs and unsafe paths fail before harness writes or deletions.

Without --yes, deinit and uninstall are dry-runs in non-interactive use; in a TTY they preview and ask for confirmation. deinit is the inverse of init for the current destination: it removes hash-matching managed harness files and unwires local Processkit MCP entries from every agent config written at init, preserves and reports modified files, safely removes only Processkit bundle keys from the shared extract registry, and forgets the destination from the ledger. Local map files are kept.

processkit uninstall can run from anywhere. It defaults to global/all: every destination in the ledger, each local Processkit MCP entry, the global Cursor MCP entry, the CLI shims and ~/.processkit. For installs created before the ledger existed:

processkit uninstall --discover ~/workspace --yes

Advanced --scope=repo|all-repos|mcp-local|mcp-global|cli|all, --project-root, and --keep-mcp filters remain available. prune remains stale-only and never performs a full deinitialization.

MCP tools

• business_process_validate
• business_impact_validate
• business_diff_scope

Processkit has no runtime dependency on CodeGraph. init preserves an existing CodeGraph MCP entry and does not require the server to be live.

Portability

Processkit never writes portable platform-repos.json / legacy-repos.json (Platform DNA / Docskit own those). It only ensures ignored platform-repos.local.json and legacy-repos.local.json skeletons so cross-repo skills have a place to resolve checkouts regardless of toolkit install order.