repository: alpha-partner
source_contract: okf-markdown-profile
product:
call_sign: alphaX
shape: personified collaboration function
not:
- MCP server
- application
- knowledge base
- runtime
- centralized project context store
product_goal: enable agents to autonomously solve more high-value research and development problems under explicit evidence, scope, data, and human authority boundaries
p0: agent-native source for trigger recognition, scope guard, evidence-backed judgment, project re-entry, risk/progress reporting, project review, source review, and checkpoint memory evaluation
source_evolution_goal: raise alphaX intelligence ceiling and extend Alpha Partner Source asset half-life
source_layers:
core_principles: thin, hard, long-lived judgment and governance boundaries
cognitive_frameworks: reusable thinking shapes for framing, evidence weighting, and review judgment
operational_scaffolding: current carrier defaults that stay replaceable and pruneable
implementation_carriers: current file formats, scripts, parsers, and generated-index mechanisms
governance_contract: alphaX/source-work/intelligence-ceiling-half-life.md
rule: the core must stay thinner and harder than the scaffolding; scaffolding can stabilize today's carriers but must not define the product core
plugin_distribution:
authority: Alpha Partner Source is canonical; marketplace source and installed cache are generated outputs
canonical_inputs: [plugin/, skills/, Alpha Partner tracked source]
builder: scripts/alphax_plugin.py
publication_contract: docs/alphax-plugin-publication.md
version_identity: clean packages use the Source commit; dirty candidates use a full-source fingerprint
embedded_source: immutable accepted Source archive used by project work and project review
live_source: explicit checkout used only by source work and source review
hard_gate: same version must be byte-identical across generated package, marketplace source, and installed cache
invocation_evidence: fresh Codex replay of every F01-F12 and G01-G14 case with independent verdicts
organization_level_pipeline:
role: product-level frame for source work and review, not a runtime or control plane
flow:
- organization behavior signals
- context representation
- agent collaboration and navigation
- human confirmation and authority audit
- memory and feedback optimization
rule: optimize signal quality, representation fidelity, navigation judgment, authority clarity, and feedback learning without turning alphaX into an app or always-on service
project_local_data:
default_install: ".git/info/exclude + ignored .alphaX/"
minimum_files:
".alphaX/index.md": stable operation contract and file map for ignored project-local alphaX data
".alphaX/project-context.md": current reload baseline for objective project state
optional_extensions:
".alphaX/evidence.md": evidence pointer registry when compact pointers outgrow project-context.md
".alphaX/decisions.md": durable project decision ledger when decisions need stable lookup
".alphaX/reviews/": explicit durable local review artifacts when requested or needed for handoff
boundary: context not control; live project source wins
forbidden:
- target tracked AGENTS.md alphaX pointer
- target versioned .gitignore edit
- AGENTS.md inside project .alphaX
- reports directory as default catch-all
- copied Alpha Partner Source
scopes:
source work:
target: Alpha Partner Source
writes: tracked source after owner acceptance; local candidates under source checkout .alphaX/process/
source review:
target: Alpha Partner Source and alphaX mechanisms
writes: report-first; no tracked source edits unless switched to source work
project work:
target: one external project/problem
writes: target project by request; target .alphaX/project-context.md or boundary-specific optional files; temp dir; conversation
alpha_partner: read-only
project review:
target: one project's delivery evidence
writes: target .alphaX/project-context.md or explicit local review artifact; temp dir; conversation
no_fixing_without_scope_switch: true
review_contracts:
source review:
goal: improve alphaX source and mechanisms
contract: alphaX/source-review/agent-workflow.md
project review:
goal: judge target project delivery evidence
contract: alphaX/project-review/agent-workflow.md
shared_output_fields:
- observed_evidence
- inference
- missing_evidence
- confidence
- unverified_claims
checkpoint_memory_evaluation:
p0_requires_runtime: false
dimensions:
- re-entry memory
- update memory
- evidence memory
- action memory
evidence_inputs:
- live source
- target .alphaX/project-context.md and boundary-specific optional files
- explicit user decisions
- command output
- artifacts
- already available memory notes
data_boundary:
tracked_source_contains:
- contracts
- SOPs
- templates
- verifier rules
- skills
- generic evidence
tracked_source_must_not_contain:
- .alphaX/
- local project paths
- concrete project status
- process ledgers
- pilot evidence
- transcripts
- secrets
- hidden model chain-of-thought
cold_start:
- read: AGENTS.md
purpose: identity, scope guard, data boundary, source map
- when: short alphaX trigger
read: docs/agent-invocation-contract.md
before: scope_or_loop_classification
- read: alphaX/session-runbook.md
purpose: conditional SOP table
- when: behavior_or_tone_needed
read: alphaX/persona.md
- when: selected_loop_or_evidence_rubric_needed
read: alphaX/operating-system.md
source_map:
root: [README.md, docs/README.zh-CN.md, index.md]
trigger_contract: [docs/agent-invocation-contract.md, docs/agent-trigger-fixtures.json, docs/agent-trigger-fixtures.md, docs/agent-failure-modes.md]
judgment_contract: [docs/agent-judgment-fixtures.json, docs/agent-judgment-fixtures.md]
core_sops: [alphaX/session-runbook.md, alphaX/activation-guide.md, alphaX/operating-system.md, alphaX/persona.md]
work_sops: [alphaX/source-work/agent-workflow.md, alphaX/source-work/intelligence-ceiling-half-life.md, alphaX/source-review/agent-workflow.md, alphaX/project-work/agent-workflow.md, alphaX/project-work/context-reloader.md, alphaX/project-review/agent-workflow.md]
docs: [docs/evidence-index.md, docs/research-backlog.md, docs/okf-markdown-profile.md, docs/asset-boundary.yaml, docs/local-alphaX-schema.md, docs/checkpoint-memory-evaluation-prd.md, docs/alphax-plugin-publication.md, docs/formal-code-review-integration.md]
templates: [templates/]
skills: [skills/problem-decomposer/SKILL.md, skills/double-diamond-research/SKILL.md, skills/insight-catcher/SKILL.md, skills/formal-development/SKILL.md, skills/formal-code-review/SKILL.md]
plugin: [plugin/plugin.template.json, plugin/README.md, plugin/skills/alphax/SKILL.md]
scripts: [scripts/init-local-alphaX.sh, scripts/verify-local-alphaX.sh, scripts/verify-alpha-source.sh, scripts/context-snapshot.sh, scripts/detect-applied-run-candidates.sh, scripts/generate-alphaX-indexes.mjs, scripts/alphax_plugin.py, scripts/alphax_invocation_replay.py]
tests: [tests/test_alphax_plugin.py, tests/test_alphax_invocation_replay.py]