pqrst-practice/AGENTS.md

pqrst-practice — Agent Instructions

Repo Identity

Purpose: Retrospective PQRST estimation practice for closing agentic coding sessions.

Domain: agents Repo slug: pqrst-practice Topic ID: c1d199b6-55ee-4db6-b49e-257a9f0f15ac Workplan prefix: PQRST-WP-

---

State Hub Integration

The Custodian State Hub tracks work across all domains. Codex uses HTTP REST and the statehub CLI by default. MCP is opt-in because the current Codex MCP bridge adds severe call latency; the full administrative MCP surface remains available to clients that need it.

Context	URL
Local workstation	http://127.0.0.1:8000
Remote via tunnel	http://127.0.0.1:18000
Optional local edge relay	http://127.0.0.1:18080

When an operator has enabled the edge relay, set API_BASE to the relay URL. Queueable writes return an explicit queued receipt if the central hub is unreachable. Treat that as pending local evidence, then ask the operator to run statehub outbox status/replay after connectivity returns.

Codex workspace-write sandboxes need network access enabled to reach the host's loopback listener. Bootstrap this once with make -C ~/state-hub configure-codex and restart Codex. The canonical REST health endpoint is /state/health, not /health. If a sandboxed loopback probe fails, retry it with escalated execution before declaring State Hub unavailable; a managed Codex permission profile may still enforce isolated networking. Experimental MCP can be enabled explicitly with make -C ~/state-hub configure-codex WITH_MCP=1.

Orient at session start

# Offline brief — works without hub connection
cat .custodian-brief.md

# Active workplans for this domain
curl -s "http://127.0.0.1:8000/workplans/?topic_id=c1d199b6-55ee-4db6-b49e-257a9f0f15ac&status=active" \
  | python3 -m json.tool

# Check inbox
curl -s "http://127.0.0.1:8000/messages/?to_agent=pqrst-practice&unread_only=true" \
  | python3 -m json.tool

Mark a message read:

curl -s -X PATCH "http://127.0.0.1:8000/messages/<id>/read" \
  -H "Content-Type: application/json" -d '{}'

Log progress (required at session close)

curl -s -X POST http://127.0.0.1:8000/progress/ \
  -H "Content-Type: application/json" \
  -d '{
    "summary": "what was done",
    "event_type": "note",
    "author": "codex",
    "workplan_id": "<uuid>",
    "task_id": "<uuid>"
  }'

Omit workplan_id / task_id when not applicable.

Update task status

curl -s -X PATCH "http://127.0.0.1:8000/tasks/<task_id>" \
  -H "Content-Type: application/json" \
  -d '{"status": "progress"}'
# values: wait | todo | progress | done | cancel

Flag a task for human review

curl -s -X PATCH "http://127.0.0.1:8000/tasks/<task_id>" \
  -H "Content-Type: application/json" \
  -d '{"needs_human": true, "intervention_note": "reason"}'

---

Session Protocol

Start:

1. cat .custodian-brief.md — domain goal and open workplans (offline-safe)
2. Check inbox: GET /messages/?to_agent=pqrst-practice&unread_only=true; mark read
3. Scan workplans: ls workplans/ — note status: ready, active, or blocked files and open tasks
4. Check human-needed tasks: GET /tasks/?needs_human=true

During work:

• Update task statuses in workplan files as tasks progress
• Record significant decisions via POST /decisions/

Close:

1. Update workplan file task statuses to reflect progress
2. If finishing a workplan: hand off residuals as live work records first (intake with origin: residual + origin_ref: <WP-id>, or a next workplan / decision / engagement). Do not park leftovers only in prose or SCOPE.md. Canon: the-custodian/canon/standards/work-record-types_v0.1.md § Residuals.
3. Log: POST /progress/ with a summary of what changed (name handoff ids)
4. After workplan file changes, run:

   uv run --project ~/repo-manager rmgr sync --path . --push
   

   This assigns only missing deterministic identifiers, verifies the pushed Forgejo commit and primary/railliance01, then requests one central reconciliation. A queued receipt is pending evidence; rerun after connectivity returns. Use statehub fix-consistency for a separate deep audit.

---

Credential and access routing

Audience: Codex, Claude Code, Grok, and custodian agents that call llm-connect for inference. Run this check before requesting secrets, API keys, SSH access, login tokens, or database passwords — in any repo, not only ops-warden.

ops-warden issues SSH certificates only (warden sign, cert_command). Every other credential need belongs to another subsystem. Do not message ops-warden on State Hub expecting a secret value; the reply is a pointer, not a key.

Lookup (do this first)

warden route find "<describe your need>" --json
warden route show <catalog-id> --json

Requires the warden CLI from ~/ops-warden (uv tool install . or uv run warden).

Agent runtime	How to orient
Codex / Grok (shell, HTTP State Hub)	warden route commands above; inbox to_agent=pqrst-practice is for coordination, not secret vending
Claude Code (MCP when available)	get_domain_summary("custodian") for workplans; still use warden route for credential ownership
llm-connect (inference service)	Never put secret retrieval in prompts; route custody to OpenBao/operator paths surfaced by warden route

Quick routing table

I need…	Owner	ops-warden executes?
SSH cert (adm/agt/atm)	ops-warden	Yes — warden sign
API key, DB password, provider token	OpenBao (railiance-platform)	No — route only
Login / OIDC / MFA	key-cape / Keycloak	No — route only
Authorization decision	flex-auth	No — route only
activity-core → issue-core emission	activity-core + issue-core	No — warden route show activity-core-issue-sink
SSH tunnel	ops-bridge (+ cert_command from warden)	No — route only

Anti-patterns (do not do these)

• POST /messages/ to ops-warden asking for ISSUE_CORE_API_KEY, OPENROUTER_API_KEY, etc.
• Inventing warden secret, warden login, warden bao, warden tunnel — they do not exist
• Pasting secrets into Git, State Hub, workplans, logs, or chat

Other capabilities (reuse-surface)

Non-credential capabilities are usually discovered through reuse-surface federation (reuse-surface registry / capability.* indexes). Credential routing is inlined in every repo's agent instructions because it is high-frequency, high-risk, and easy to get wrong.

Canon: ~/ops-warden/wiki/CredentialRouting.md · catalog ~/ops-warden/registry/routing/catalog.yaml

---

Workplan Convention (ADR-001)

Work items originate as files in this repo — not in the hub. The hub is a read/cache/index layer that rebuilds from files.

File location: workplans/PQRST-WP-NNNN-<slug>.md

Archived location: finished workplans may move to workplans/archived/YYMMDD-PQRST-WP-NNNN-<slug>.md. The YYMMDD prefix is the completion/archive date; the frontmatter id does not change.

Ad Hoc Tasks: small opportunistic fixes discovered during a session use workplans/ADHOC-YYYY-MM-DD.md, workplan id PQRST-WP-ADHOC-YYYY-MM-DD, and task ids PQRST-WP-ADHOC-YYYY-MM-DD-T01, etc. PQRST-WP includes its final -WP token. Unqualified historic ADHOC-* ids are grandfathered and must not be copied into new records. Use this only for low-risk work completed directly; create a normal workplan for anything needing analysis, design, approval, dependencies, or multiple phases.

Frontmatter:

---
id: PQRST-WP-NNNN
type: workplan
title: "..."
domain: agents
repo: pqrst-practice
status: proposed | ready | active | blocked | backlog | finished | archived
owner: codex
topic_slug: ...
created: "YYYY-MM-DD"
updated: "YYYY-MM-DD"
state_hub_workstream_id: "<uuid>"   # deterministic UUIDv5; managed by Repo Manager
---

Use proposed for a new draft, ready after review against current repo state, and finished after implementation. stalled and needs_review are derived health labels, not frontmatter statuses.

Terminology: workplan is the fleet term; workstream appears only in legacy API/MCP/frontmatter bridges until STATE-WP-0069 retires them — see the-custodian/canon/standards/workplan-terminology-fleet_v0.1.md.

Task block format (one per ## section):

## Task Title

` ` `task
id: PQRST-WP-NNNN-T01
status: wait | todo | progress | done | cancel
priority: high | medium | low
state_hub_task_id: "<uuid>"         # deterministic UUIDv5; managed by Repo Manager
` ` `

Task description text.

Status progression: todo → progress → done; use wait for waiting/blocked work and cancel for stopped work.

Residuals when finishing: actionable leftovers become live work records before status: finished — usually an intake (origin: residual, origin_ref: PQRST-WP-NNNN) or a spawned workplan. Residual is a role, not a kind. Fleet list lives on State Hub, not in SCOPE.md.

To create a new workplan:

1. Write the file following the format above
2. Run uv run --project ~/repo-manager rmgr sync --path . --push.
3. Run statehub fix-consistency only when a separate deep audit is needed.

---

Repo-Specific Working Notes

What this repo is

A specification-and-prompt repository. Every tracked file is Markdown except .repo-classification.yaml. There is no application code, no package manifest, and no runtime. Read INTENT.md and SCOPE.md before changing anything — SCOPE.md is normative about what must not be added here.

The deliverables are:

File	Role
spec/PqrstEstimationPractice.md	The normative specification. Versioned.
PqrstPrompt.md	The canonical end-of-session prompt operators paste.
INTENT.md / SCOPE.md	Why the practice exists; what belongs here.
history/	Source drafts. Append-only — never edit an existing file.

Developer workflow

There is nothing to install, build, or run.

Step	Command	Notes
Install	—	No dependencies.
Build	—	No build step.
Run	—	Nothing executes; the prompt is pasted into a session by an operator.
Test	—	No test suite. Verification is the checklist below.
Lint	—	No linter is configured. Match surrounding Markdown style.
Sync	uv run --project ~/repo-manager rmgr sync --path . --push	After workplan file changes.
Audit	statehub fix-consistency	Deep audit; also regenerates WORK-RECORDS.md.

Do not add a toolchain to make this table look fuller. A validator, collector, or CI harness for PQRST records is explicitly out of scope (SCOPE.md) and belongs in a consuming repository.

Verification checklist

Run this before committing a change to the spec or the prompt. It is a reading task, not a command — the coupling between the two files is the thing that breaks.

1. Prompt matches spec. Every rule in the prompt's Rules: list has a corresponding rule in spec §3, and the prompt's output block matches the stored-record format in spec §5.1 field for field.

2. The 5-tuple is closed. No change introduces a sixth dimension, and confidence stays outside P/Q/R/S/T (spec R9, R10).

3. Examples validate. Every example record in the repo satisfies spec §5.5. The arithmetic can be checked mechanically:

   grep -rhoE 'P[0-9]+ Q[0-9]+ R[0-9]+ S[0-9]+ T[0-9]+' --include='*.md' . \
     | sort -u \
     | awk '{sig=$0; s=0; for(i=1;i<=NF;i++){gsub(/[PQRST]/,"",$i); s+=$i}
            printf "%-28s %s\n", sig, (s==100 ? "ok" : "SUM=" s)}'
   

   Every line must report ok.

4. Cross-references resolve. Relative links between INTENT.md, SCOPE.md, PqrstPrompt.md, and spec/ still point at existing files and sections.

5. Version discipline. A change to a dimension's meaning, the validation rules, or the stored record format is breaking: bump the spec version, add a row to its Appendix B, and say plainly that older records are less comparable. Editorial changes need no bump.

6. history/ untouched. git diff --stat history/ is empty.

Conventions

• The specification is the source of truth; PqrstPrompt.md follows it, never the reverse.
• Prefer prose the operator can act on at 11pm at the end of a long session. Terse beats thorough in PqrstPrompt.md; thorough beats terse in spec/.
• Keep the anti-goals in INTENT.md intact when editing. They are the load- bearing part of the practice, not framing.