Docs project management

TASKLANG TYPE SKILL

IDENTITY "Docs Project Management (the docs/-native system of record)"

A universal, reusable playbook for running a whole project — requirements, planning, build tracking, testing, bugs, and team reporting — as committed markdown inside the repo's docs/ folder. It works in ANY project, in any stack, from day one of a NEW project: no external issue tracker, no ticket system, no proprietary tooling. A work item is a markdown file with YAML frontmatter; a relationship is a relative markdown link; a status change is a frontmatter edit; the full history is git.

It does three jobs:

1. WORK-ITEM MODEL — WHAT document types exist (PRD, TRD, Epic, User Story, Test Case, Bug, Roadmap), WHERE each lives, HOW they link, and HOW status is tracked.
2. TEAM WORK-REPORTING — HOW every agent/contributor reports its work so anyone can see, in one committed place, everything that was done (who, what, when, with which evidence).
3. DURABLE vs TEMPORARY — WHAT belongs in committed docs/ versus the git-ignored .ai_log/ temp-evidence folder, and when to promote content from one to the other.

---

§1 Document types — WHAT the system manages

Every work item is a committed markdown file in its docs/ subfolder: descriptive kebab-case filename (never an opaque ID), YAML frontmatter for status/metadata, relative-markdown links for relationships. Full field-level detail: [[work-item-taxonomy]].

TABLE document-types COLUMNS: Type, Lives in, What it is / holds ROW: Project brief | docs/project_brief.md | The product spec — the SOURCE OF TRUTH the whole project derives from; wins over any assumption ROW: Project config| docs/project_config/info.md | The single home for project-specific FACTS (product name, app name, stack, domain, scope, success scenarios) — agents read it instead of hard-coding facts ROW: Roadmap | docs/ROADMAP.md + docs/roadmap.json | The milestone/release plan (human file) + a machine-readable MIRROR of every work item's current status (JSON cache, regenerated from docs/, committed) ROW: PRD | docs/prd/ | Product Requirements Document — WHAT to build and WHY: objectives, in-scope / out-of-scope, testable acceptance criteria per success scenario ROW: TRD | docs/trd/ | Technical Requirements Document — HOW to build it: architecture, data model, migrations, jobs, authz. Canonical TRD home; keep index.json (machine registry, one entry per TRD) + index.md (human index) beside the files ROW: Epic | docs/epics/ | A large feature/theme grouping related User Stories; carries goal + scope + its story list ROW: User Story | docs/user_stories/ | The smallest shippable requirement — "as a <user> I want <capability>" + numbered acceptance criteria; filename NNN-<kebab-title>.md ROW: Test Case | docs/test_cases/ | A verification script for ONE behavior — numbered navigate→action→expected steps; filename tc-NNN-<kebab-behavior>.md; MUST follow the [[test-case-template]] and link its story "is tested by" ROW: Bug | docs/bugs/ | A defect AND its Root-Cause-Analysis in ONE file — Symptom/Reproduction + 5-Whys RCA + Preventative Action at detection; Fix/diff/Resolved sections completed by whoever fixes it ([[bug-rca-report]]); blocks its story ROW: Session report| docs/sessions/<YYYY-MM-DD>-<slug>/ | Per-run team work folder — a README manifest + one bookended report per agent/phase ([[session-reporting]]) ROW: Feature log | docs/reports/feature-log.md + feature-log/ | The team work-log split by feature: one lean index row per feature, linking a per-feature file where each run appends a row

RULES - Filenames: descriptive kebab-case, never opaque IDs. External refs (ticket numbers), when they exist, go in frontmatter + the title line — not the filename. - Each subfolder may carry a template_<type>.md seed; new items copy its shape. - Only APPROVED/final content advances status; drafts stay in their lifecycle state.

---

§2 The link graph — HOW documents relate

Every relationship is a labelled RELATIVE-MARKDOWN link in a ## Linked Documents section, written in BOTH files (add the reciprocal link). No link-type IDs, no tracker mechanics — the label carries the semantic. Full table + exact syntax: [[link-graph]].

• User Story is part of its Epic; the Epic lists it back (realized by).
• Test Case ↔ User Story is ALWAYS "is tested by" (story side) / "tests" (test-case side) — never "relates to" or "part of".
• Bug blocks the User Story it breaks; the story links the bug back.
• PRD ↔ TRD cross-link (specified by); User Story is derived from its PRD.

---

§3 Status lifecycle — HOW progress is tracked

Status lives in each file's YAML frontmatter status: field; changing status = editing that field. Full lifecycle, required frontmatter fields, and the signature convention: [[status-lifecycle]].

LIFECYCLE status VALUES: backlog → ready → in_progress → in_review → qa → done (blocked = orthogonal flag)

Pick the closest value per item type: a Bug walks open → in_progress → fixed → verified → closed; a Test Case walks draft → ready → automated.

BOOKEND both-ends

• ON START: set status: in_progress BEFORE touching the work; stamp updated:.
• ON COMPLETION: set the next appropriate state (in_review / qa / done); stamp updated:.
• NEVER leave a worked file on its old status — that is an incomplete bookend.

SIGNATURE

• Sign every file update / report entry: — <agent-name> v<version> (<passport-id>).

---

§4 Team work-reporting — HOW work becomes visible

One committed place per run where anyone can see all work done. Full layout, report schema, and per-role sections: [[session-reporting]].

1. The orchestrator (or the first agent of a run) CREATES the session folder docs/sessions/<YYYY-MM-DD>-<slug>/ with a README.md manifest (request, plan, a phase→agent→status→report table) and passes session_dir to every worker.
2. EVERY agent bookends its run there — <session_dir>/<phase-N>-<agent>.md, created with a STARTING entry, updated on COMPLETION (work done, work items touched as links, evidence links), flipping its row in the session README.
3. Each agent ALSO appends a run-row to its feature's log docs/reports/feature-log/<slug>.md (indexed by the lean docs/reports/feature-log.md).
4. Reports carry the role-specific section: QA → coverage report; security → findings + verdict; devops → deploy/CI status; architect → gate verdict; PM → acceptance verdict.

---

§5 docs/ vs .ai_log/ — durable vs temporary

The one split that keeps context lean AND the record complete. Never confuse the two.

docs/ — durable, committed, the system of record. Anything with lasting value lives here: specs, decisions, work items, reports, audits, backlog/deferred-work notes, coverage summaries.

.ai_log/ — temporary, git-ignored, evidence offload ONLY. The folder is tracked via .gitkeep; its CONTENTS are ignored and therefore never committed — anything left only there is effectively lost. Belongs in .ai_log/:

• screenshots / diff images / browser captures taken as run evidence
• run logs, test-run output, scratch files, intermediate payloads too big for context
• the per-session pipeline-state handoff JSON (.ai_log/session-<id>-<name>.json) — the machine bus that carries key fields + PATHS between phases
• naming: .ai_log/<phase-N>-<agent>-<artifact>.<ext>

RULES - Hand off by PATH, never by inlining a blob into a prompt or a state file. - PROMOTE anything durable that lands in .ai_log/ into the right docs/ subfolder (with frontmatter + links) before relying on it — a decision or report parked only in .ai_log/ does not exist as a record. - Reports in docs/ may LINK to .ai_log/ evidence, knowing it is ephemeral.

---

§6 Adopting in a NEW project — bootstrap

CHECKLIST bootstrap [ ] Create the skeleton: docs/{prd,trd,epics,user_stories,test_cases,bugs,sessions,reports/feature-log,project_config}/ + docs/ROADMAP.md [ ] Write docs/project_brief.md (the spec) and docs/project_config/info.md (the project facts) [ ] Create .ai_log/ with a tracked .gitkeep; git-ignore its contents (/.ai_log/* + !/.ai_log/.gitkeep) [ ] Seed docs/reports/feature-log.md (empty index) and docs/roadmap.json (empty mirror) [ ] Drop a template_<type>.md into each work-item subfolder if you want enforced shapes

Nothing else is required — the model is just folders + markdown + frontmatter + relative links.

---

The Map — load the one reference you need

This SKILL.md is a MAP. Deep detail lives in atomized references/ notes; load only what the task needs. Machine-readable nav: maps/index.json (+ tags.json, links.json, manifest.json).

TABLE references COLUMNS: Note, Owns, Load when ROW: [[work-item-taxonomy]] | §1 detail — folder, filename pattern, owner per work item | You need a work item's folder, filename pattern, or owner ROW: [[link-graph]] | §2 detail — every relationship's semantic, direction, exact syntax | Linking two work items / checking a relationship's direction ROW: [[status-lifecycle]] | §3 detail — lifecycle values, bookend, required frontmatter fields, signature | Starting/finishing work, setting status, filling frontmatter ROW: [[test-case-template]] | The mandatory Test Case Template (all fields) | Authoring/reviewing a docs/test_cases/ file ROW: [[bug-rca-report]] | The Bug+RCA format — detection sections (Symptom/Repro/5-Whys/Preventative Action) + fix sections (Fix/diff/Resolved) | Authoring/completing a docs/bugs/ Bug file ROW: [[session-reporting]] | §4 detail — session folder layout, report schema, role sections, feature log | Bookending your run as a report; organizing a session folder ROW: [[conventions-recap]] | The non-negotiables checklist + the Knowledge Strategy | A one-glance rule check; rules for updating this skill

---

Non-negotiables (one-glance)

CHECKLIST conventions [ ] Every work item = a committed markdown file in its §1 subfolder, kebab-named, frontmatter'd [ ] Every relationship = a labelled relative-markdown link, written reciprocally in both files [ ] Status bookended at BOTH ends by editing frontmatter status: — never left stale [ ] Test Cases follow the template AND link their story "is tested by" [ ] Bugs are one Bug+RCA file: detection sections at filing, fix sections completed by the fixer [ ] Every run bookends a report in docs/sessions/<session>/ + a feature-log row [ ] Durable content in docs/ (committed); ONLY temporary evidence in .ai_log/ (git-ignored); promote anything durable that lands in .ai_log/ [ ] No external tracker mechanics — everything is a local docs/ file, linked by relative paths

---

Knowledge Strategy

• Patterns to capture: recurring item shapes, link-graph edge cases, report sections that proved useful.
• Update permission: agents may freely add/update files in references/; changes to THIS SKILL.md map require user approval. Full strategy: [[conventions-recap]].